{"id":"1950-10-turing-computing-machinery-and-intelligence","date":"1950-10","precision":"month","title":"Turing asks whether machines can think, then changes the question","summary":"Turing replaced \"can machines think\" with a question you can actually run: could a machine fool an interrogator into believing it was human. Seventy-five years later we are still arguing about the answer and still using his framing.","category":"research","orgs":["University of Manchester"],"tier":1,"source":{"url":"https://academic.oup.com/mind/article/LIX/236/433/986238","title":"I.—Computing Machinery and Intelligence","publisher":"Mind","author":"A. M. Turing","type":"paper","date":"1950-10"}}
{"id":"1956-dartmouth-workshop","date":"1956","precision":"year","title":"The Dartmouth workshop names the field","summary":"Ten researchers spent a summer at Dartmouth on the premise that every feature of intelligence could be described precisely enough for a machine to simulate it. The proposal gave the field its name and its founding overconfidence.","category":"research","orgs":["Dartmouth College"],"tier":1,"source":{"url":"https://ojs.aaai.org/aimagazine/index.php/aimagazine/article/view/1904","title":"A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence, August 31, 1955","publisher":"AI Magazine","type":"paper","date":"2006"}}
{"id":"1958-07-perceptron-demonstrated","date":"1958-07","precision":"month","title":"Rosenblatt's perceptron learns in public","summary":"The US Office of Naval Research fed punched cards to a five-ton IBM 704, and after fifty tries it could tell cards marked on the left from cards marked on the right. Rosenblatt called it the first machine capable of an original idea, and the press printed that.","category":"research","orgs":["Cornell Aeronautical Laboratory"],"tier":2,"source":{"url":"https://news.cornell.edu/stories/2019/09/professors-perceptron-paved-way-ai-60-years-too-soon","title":"Professor's perceptron paved the way for AI, 60 years too soon","publisher":"Cornell Chronicle","type":"retrospective","date":"2019-09-25"}}
{"id":"1959-07-samuel-checkers","date":"1959-07","precision":"month","title":"Arthur Samuel names machine learning","summary":"Samuel's checkers program learned to play a better game than the man who wrote it, and his paper handed the field the phrase it has used ever since. The one-line definition everybody attributes to him appears nowhere in it.","category":"research","orgs":["IBM"],"tier":2,"source":{"url":"https://ieeexplore.ieee.org/document/5392560","title":"Some Studies in Machine Learning Using the Game of Checkers","publisher":"IBM Journal of Research and Development","author":"Arthur L. Samuel","type":"paper","date":"1959-07"},"references":[{"url":"https://people.csail.mit.edu/brooks/idocs/Samuel.pdf","title":"Some Studies in Machine Learning Using the Game of Checkers","publisher":"MIT CSAIL","type":"paper","date":"1959-07"}]}
{"id":"1966-01-eliza","date":"1966-01","precision":"month","title":"ELIZA proves how little it takes to be believed","summary":"Weizenbaum's script reflected your own sentences back at you as questions, and people confided in it anyway, including his secretary, who asked him to leave the room. He spent the rest of his career alarmed by what that revealed.","category":"research","orgs":["MIT"],"tier":2,"source":{"url":"https://dl.acm.org/doi/10.1145/365153.365168","title":"ELIZA—a computer program for the study of natural language communication between man and machine","publisher":"Communications of the ACM","author":"Joseph Weizenbaum","type":"paper","date":"1966-01"}}
{"id":"1973-lighthill-report","date":"1973","endDate":"1980","precision":"year","title":"The Lighthill report starts the first AI winter","summary":"A review commissioned by the UK research council concluded that AI had failed to deliver on any of its grand claims, and funding collapsed on both sides of the Atlantic. The field learned that promising general intelligence has a bill attached.","category":"culture","orgs":["Science Research Council"],"tier":2,"source":{"url":"https://www.aiai.ed.ac.uk/events/lighthill1973/","title":"The Lighthill Debate (1973)","publisher":"University of Edinburgh","type":"retrospective"}}
{"id":"1986-10-09-backpropagation-in-nature","date":"1986-10-09","precision":"day","title":"Backpropagation reaches Nature","summary":"Rumelhart, Hinton and Williams showed that a network could learn useful internal representations by propagating error backwards through its layers, and this is the paper that made the technique widely known. Earlier versions existed, credited in the paper itself, and it took another two decades before hardware caught up to train networks of any real depth.","category":"research","orgs":["Carnegie Mellon University","University of California San Diego"],"tier":1,"source":{"url":"https://www.nature.com/articles/323533a0","title":"Learning representations by back-propagating errors","publisher":"Nature","type":"paper","date":"1986-10-09"}}
{"id":"1995-09-support-vector-networks","date":"1995-09","precision":"month","title":"Support vector machines take the field","summary":"Cortes and Vapnik added a soft margin to the kernel classifiers Boser, Guyon and Vapnik had introduced three years earlier, letting the method handle data that could not be perfectly separated. For the next fifteen years, if you were doing machine learning, you were most likely doing some version of this.","category":"research","orgs":["AT&T"],"tier":2,"source":{"url":"https://link.springer.com/article/10.1007/BF00994018","title":"Support-vector networks","publisher":"Machine Learning","author":"Corinna Cortes and Vladimir Vapnik","type":"paper","date":"1995-09"}}
{"id":"1997-05-11-deep-blue-beats-kasparov","date":"1997-05-11","precision":"day","title":"Deep Blue beats Kasparov","summary":"IBM’s Deep Blue defeated reigning world chess champion Garry Kasparov in a six-game match. The deciding sixth game lasted 19 moves.","category":"culture","orgs":["IBM"],"tier":1,"source":{"url":"https://www.ibm.com/history/deep-blue","title":"Deep Blue","publisher":"IBM","type":"retrospective"}}
{"id":"1997-11-long-short-term-memory","date":"1997-11","precision":"month","title":"LSTM gives networks a memory","summary":"Hochreiter and Schmidhuber solved the vanishing gradient problem with a gated cell that could hold information across long sequences. It ran sequence modeling for two decades, until attention took the job.","category":"research","orgs":["Technische Universität München","IDSIA"],"tier":2,"source":{"url":"https://direct.mit.edu/neco/article/9/8/1735/6109/Long-Short-Term-Memory","title":"Long Short-Term Memory","publisher":"Neural Computation","type":"paper","date":"1997-11"}}
{"id":"1998-lenet-5","date":"1998","precision":"year","title":"Convolutional networks go to work reading checks","summary":"LeCun's group had a convolutional network reading millions of bank checks in production, and shipped MNIST, the handwritten digit set a generation of researchers learned on. The architecture anticipated the convolutional networks that would win ImageNet fourteen years later, still waiting on the data and hardware that made that possible.","category":"research","orgs":["AT&T","Université de Montréal"],"tier":2,"source":{"url":"https://ieeexplore.ieee.org/document/726791","title":"Gradient-based learning applied to document recognition","publisher":"Proceedings of the IEEE","author":"Yann LeCun, Léon Bottou, Yoshua Bengio and Patrick Haffner","type":"paper","date":"1998"},"references":[{"url":"https://hal.science/hal-03926082","title":"Gradient-based learning applied to document recognition","publisher":"HAL open science","type":"paper","date":"1998"}]}
{"id":"2001-10-random-forests","date":"2001-10","precision":"month","title":"Random forests make the ensemble the default","summary":"Breiman combined decision trees trained on random samples of observations and features, then used their votes for classification. His analysis showed that adding trees did not keep increasing generalization error.","category":"research","orgs":["University of California Berkeley"],"tier":3,"source":{"url":"https://link.springer.com/article/10.1023/A:1010933404324","title":"Random Forests","publisher":"Machine Learning","author":"Leo Breiman","type":"paper","date":"2001-10"},"references":[{"url":"https://www.stat.berkeley.edu/~breiman/randomforest2001.pdf","title":"Random Forests","publisher":"UC Berkeley Department of Statistics","type":"paper","date":"2001-10"}]}
{"id":"2006-07-deep-belief-nets","date":"2006-07","precision":"month","title":"Hinton makes deep networks trainable again","summary":"Hinton, Osindero and Teh trained a deep network one layer at a time. Their untuned network beat a plain backpropagation network and matched tuned methods on handwritten-digit recognition.","category":"research","orgs":["University of Toronto"],"tier":2,"source":{"url":"https://direct.mit.edu/neco/article/18/7/1527/7065/A-Fast-Learning-Algorithm-for-Deep-Belief-Nets","title":"A Fast Learning Algorithm for Deep Belief Nets","publisher":"Neural Computation","author":"Geoffrey E. Hinton, Simon Osindero and Yee-Whye Teh","type":"paper","date":"2006-07"},"references":[{"url":"https://www.cs.toronto.edu/~hinton/absps/fastnc.pdf","title":"A fast learning algorithm for deep belief nets","publisher":"University of Toronto","type":"paper","date":"2006-07"}]}
{"id":"2009-06-imagenet","date":"2009-06","precision":"month","title":"ImageNet turns data into the bottleneck","summary":"Fei-Fei Li’s group used paid crowd workers to label millions of images over two years, building a dataset that became the basis for the ImageNet recognition challenge.","category":"research","orgs":["Princeton University"],"tier":2,"source":{"url":"https://ieeexplore.ieee.org/document/5206848","title":"ImageNet: A large-scale hierarchical image database","publisher":"IEEE CVPR","type":"paper","date":"2009-06"}}
{"id":"2009-09-21-netflix-prize","date":"2009-09-21","precision":"day","title":"The Netflix Prize pays out $1 million","summary":"Three years of open competition ended with an ensemble of blended models beating Netflix's own recommender by ten percent. Netflix paid the million, never shipped the winning system, and cancelled the sequel after researchers de-anonymized the contest data.","category":"culture","orgs":["Netflix"],"tier":3,"source":{"url":"https://archive.nytimes.com/bits.blogs.nytimes.com/2009/09/21/netflix-awards-1-million-prize-and-starts-a-new-contest/","title":"Netflix Awards $1 Million Prize and Starts a New Contest","publisher":"The New York Times","type":"news","date":"2009-09-21"},"references":[{"url":"https://www.wired.com/2010/03/netflix-cancels-contest/","title":"NetFlix Cancels Recommendation Contest After Privacy Lawsuit","publisher":"Wired","type":"news","date":"2010-03-12"}]}
{"id":"2011-02-16-watson-wins-jeopardy","date":"2011-02-16","precision":"day","title":"Watson wins Jeopardy","summary":"IBM's question-answering system beat Ken Jennings and Brad Rutter on television, then spent a decade failing to become a business. The demo was real and the product never was.","category":"culture","orgs":["IBM"],"tier":2,"source":{"url":"https://www.ibm.com/history/watson-jeopardy","title":"Watson and Jeopardy!","publisher":"IBM","type":"retrospective"}}
{"id":"2011-10-04-siri","date":"2011-10-04","precision":"day","title":"Apple puts a voice assistant in everyone's pocket","summary":"Apple bought Siri in 2010 and shipped it as the headline feature of the iPhone 4S. For the next decade \"AI assistant\" meant a thing that set timers and misheard names, which is a fair part of why ChatGPT landed the way it did.","category":"product","orgs":["Apple"],"tier":2,"source":{"url":"https://www.apple.com/newsroom/2011/10/04Apple-Launches-iPhone-4S-iOS-5-iCloud/","title":"Apple Launches iPhone 4S, iOS 5 & iCloud","publisher":"Apple","type":"press-release","date":"2011-10-04"},"references":[{"url":"https://www.nbcnews.com/news/world/meet-siri-your-personal-iphone-assistant-flna120221","title":"Meet Siri, your personal iPhone assistant","publisher":"NBC News","type":"news","date":"2011-10-04"}]}
{"id":"2012-06-26-google-brain-cat-neuron","date":"2012-06-26","precision":"day","title":"Google finds a cat neuron in YouTube","summary":"A network trained on unlabeled YouTube frames developed a feature responsive to cat faces. The experiment used 16,000 cores and about a billion connections to study unsupervised learning at scale.","category":"research","orgs":["Google"],"tier":2,"source":{"url":"https://blog.google/innovation-and-ai/products/using-large-scale-brain-simulations-for/","title":"Using large-scale brain simulations for machine learning and A.I.","publisher":"Google","type":"announcement","date":"2012-06-26"},"references":[{"url":"https://arxiv.org/abs/1112.6209","title":"Building high-level features using large scale unsupervised learning","publisher":"arXiv","type":"paper","date":"2011-12-29"}]}
{"id":"2012-12-alexnet","date":"2012-12","precision":"month","title":"AlexNet wins ImageNet and everyone changes jobs","summary":"Krizhevsky, Sutskever and Hinton nearly halved the ImageNet error rate with a convolutional network trained on two consumer GPUs, demonstrating a large improvement on the competition benchmark.","category":"research","orgs":["University of Toronto"],"tier":1,"source":{"url":"https://papers.nips.cc/paper_files/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html","title":"ImageNet Classification with Deep Convolutional Neural Networks","publisher":"NeurIPS","type":"paper","date":"2012-12"}}
{"id":"2013-01-16-word2vec","date":"2013-01-16","precision":"day","title":"word2vec turns words into arithmetic","summary":"Mikolov’s group at Google introduced efficient methods for training word vectors on billions of words. The learned vectors supported analogies, including the relation between man and woman and between king and queen.","category":"research","orgs":["Google"],"tier":2,"source":{"url":"https://arxiv.org/abs/1301.3781","title":"Efficient Estimation of Word Representations in Vector Space","publisher":"arXiv","type":"paper","date":"2013-01-16"},"references":[{"url":"https://arxiv.org/abs/1310.4546","title":"Distributed Representations of Words and Phrases and their Compositionality","publisher":"arXiv","type":"paper","date":"2013-10-16"}]}
{"id":"2014-01-26-google-acquires-deepmind","date":"2014-01-26","precision":"day","title":"Google acquires DeepMind","summary":"Google paid a reported £400 million for a London lab with about fifty employees and no product. The price looked absurd at the time and then, gradually, did not.","category":"acquisition","orgs":["Google","DeepMind"],"tier":1,"source":{"url":"https://www.theguardian.com/technology/2014/jan/27/google-acquires-uk-artificial-intelligence-startup-deepmind","title":"Google acquires UK artificial intelligence startup DeepMind","publisher":"The Guardian","type":"news","date":"2014-01-27"}}
{"id":"2014-06-10-generative-adversarial-networks","date":"2014-06-10","precision":"day","title":"Generative adversarial networks arrive","summary":"Goodfellow and collaborators introduced a training setup with a generator and a discriminator. The two networks improved against one another and produced synthetic images in the paper’s experiments.","category":"research","orgs":["Université de Montréal"],"tier":2,"source":{"url":"https://arxiv.org/abs/1406.2661","title":"Generative Adversarial Networks","publisher":"arXiv","type":"paper","date":"2014-06-10"}}
{"id":"2014-11-06-amazon-echo","date":"2014-11-06","precision":"day","title":"Amazon puts Alexa in a speaker","summary":"Amazon introduced Echo as an invitation-only voice assistant in a home speaker. Its always-listening microphone made voice interaction available in an ordinary living room.","category":"product","orgs":["Amazon"],"tier":2,"source":{"url":"https://www.cnbc.com/2014/11/06/amazon-surprises-with-new-device-a-voice-assistant.html","title":"Amazon surprises with new device, a voice assistant","publisher":"CNBC","type":"news","date":"2014-11-06"},"references":[{"url":"https://www.theverge.com/2014/11/6/7167793/amazon-echo-speaker-announced","title":"Amazon just surprised everyone with a crazy speaker that talks to you","publisher":"The Verge","type":"news","date":"2014-11-06"},{"url":"https://www.amazon.science/blog/alexa-at-five-looking-back-looking-forward","title":"Alexa at five: Looking back, looking forward","publisher":"Amazon Science","type":"retrospective","date":"2019-11-06"}]}
{"id":"2015-11-09-tensorflow-open-sourced","date":"2015-11-09","precision":"day","title":"Google open sources TensorFlow","summary":"Google released its internal machine learning framework and set the default toolchain for the next five years. Owning the ecosystem turned out to be worth considerably more than keeping the code.","category":"open-source","orgs":["Google"],"tier":2,"source":{"url":"https://research.google/blog/tensorflow-googles-latest-machine-learning-system-open-sourced-for-everyone/","title":"TensorFlow: Google's latest machine learning system, open sourced for everyone","publisher":"Google Research","type":"announcement","date":"2015-11-09"}}
{"id":"2015-12-10-resnet","date":"2015-12-10","precision":"day","title":"ResNet makes depth trainable","summary":"Microsoft Research let each block learn a residual instead of a whole mapping, and networks that had been getting worse past twenty layers trained cleanly at a hundred and fifty. The skip connection is now in every transformer on this page.","category":"research","orgs":["Microsoft Research"],"tier":2,"source":{"url":"https://arxiv.org/abs/1512.03385","title":"Deep Residual Learning for Image Recognition","publisher":"arXiv","type":"paper","date":"2015-12-10"},"references":[{"url":"https://www.image-net.org/challenges/LSVRC/2015/results.php","title":"ILSVRC2015 Results","publisher":"ImageNet","type":"dataset"}]}
{"id":"2015-12-11-openai-founded","date":"2015-12-11","precision":"day","title":"OpenAI is founded as a nonprofit","summary":"A billion dollars in pledges, a research lab, and a mandate to make sure artificial general intelligence benefited all of humanity rather than any one company. Every clause of that sentence has since been litigated.","category":"company","orgs":["OpenAI"],"tier":1,"source":{"url":"https://openai.com/index/introducing-openai/","title":"Introducing OpenAI","publisher":"OpenAI","type":"announcement","date":"2015-12-11"}}
{"id":"2016-03-15-alphago-beats-lee-sedol","date":"2016-03-15","precision":"day","title":"AlphaGo beats Lee Sedol","summary":"DeepMind's system took four of five games in Seoul, including a move that professionals first called a mistake and later called beautiful. Go had been the polite example of what machines could not do.","category":"research","orgs":["DeepMind"],"tier":1,"source":{"url":"https://deepmind.google/research/alphago/","title":"AlphaGo","publisher":"Google DeepMind","type":"retrospective"},"references":[{"url":"https://www.nature.com/articles/nature16961","title":"Mastering the game of Go with deep neural networks and tree search","publisher":"Nature","type":"paper","date":"2016-01-27"}]}
{"id":"2016-03-23-microsoft-tay","date":"2016-03-23","precision":"day","title":"Microsoft's Tay lasts sixteen hours","summary":"Microsoft released a chatbot that learned from whoever talked to it, and Twitter taught it to praise Hitler before the day was out. It is still the standard citation for shipping a model without thinking about who is on the other end.","category":"culture","orgs":["Microsoft"],"tier":2,"source":{"url":"https://www.theguardian.com/technology/2016/mar/24/tay-microsofts-ai-chatbot-gets-a-crash-course-in-racism-from-twitter","title":"Tay, Microsoft's AI chatbot, gets a crash course in racism from Twitter","publisher":"The Guardian","type":"news","date":"2016-03-24"},"references":[{"url":"https://blogs.microsoft.com/blog/2016/03/25/learning-tays-introduction/","title":"Learning from Tay's introduction","publisher":"Microsoft","type":"retrospective","date":"2016-03-25"},{"url":"https://arstechnica.com/information-technology/2016/03/microsoft-terminates-its-tay-ai-chatbot-after-she-turns-into-a-nazi/","title":"Microsoft terminates its Tay AI chatbot after she turns into a Nazi","publisher":"Ars Technica","type":"news","date":"2016-03-24"}]}
{"id":"2017-06-12-attention-is-all-you-need","date":"2017-06-12","precision":"day","title":"Attention Is All You Need introduces the transformer","summary":"The Transformer used attention to connect words directly, making translation-model training more parallel. Its encoder and decoder became starting points for BERT and GPT, which adapted the architecture to learn from unlabeled text.","category":"research","orgs":["Google"],"tier":1,"source":{"url":"https://arxiv.org/abs/1706.03762","title":"Attention Is All You Need","publisher":"arXiv","type":"paper","date":"2017-06-12"}}
{"id":"2017-10-18-alphago-zero","date":"2017-10-18","precision":"day","title":"AlphaGo Zero learns Go from nothing","summary":"DeepMind reported that AlphaGo Zero trained entirely through self-play, without human game data, and beat the earlier AlphaGo version 100 games to zero.","category":"research","orgs":["DeepMind"],"tier":2,"source":{"url":"https://deepmind.google/discover/blog/alphago-zero-starting-from-scratch/","title":"AlphaGo Zero: Starting from scratch","publisher":"Google DeepMind","type":"announcement","date":"2017-10-18"},"references":[{"url":"https://www.nature.com/articles/nature24270","title":"Mastering the game of Go without human knowledge","publisher":"Nature","type":"paper","date":"2017-10-19"}]}
{"id":"2017-12-05-alphazero","date":"2017-12-05","precision":"day","title":"AlphaZero generalizes self-play beyond Go","summary":"The same algorithm, given only the rules, taught itself chess and shogi in hours and beat the strongest engines in both. The chess world spent a year arguing about the match conditions and then rebuilt its engines around the method.","category":"research","orgs":["DeepMind"],"tier":3,"source":{"url":"https://arxiv.org/abs/1712.01815","title":"Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm","publisher":"arXiv","type":"paper","date":"2017-12-05"},"references":[{"url":"https://deepmind.google/discover/blog/alphazero-shedding-new-light-on-chess-shogi-and-go/","title":"AlphaZero: Shedding new light on chess, shogi, and Go","publisher":"Google DeepMind","type":"announcement","date":"2018-12-06"}]}
{"id":"2018-05-08-google-duplex","date":"2018-05-08","precision":"day","title":"Google Duplex books a haircut over the phone","summary":"An assistant called a salon, said \"mm-hmm\" in the right places, and made an appointment without the person on the line appearing to notice. The demo drew a week of argument about disclosure and Google conceded that the calls would identify themselves.","category":"product","orgs":["Google"],"tier":3,"source":{"url":"https://research.google/blog/google-duplex-an-ai-system-for-accomplishing-real-world-tasks-over-the-phone/","title":"Google Duplex: An AI System for Accomplishing Real-World Tasks Over the Phone","publisher":"Google Research","type":"announcement","date":"2018-05-08"},"references":[{"url":"https://www.theverge.com/2018/5/10/17342414/google-duplex-ai-assistant-voice-calls-ethical-social-implications","title":"Google now says controversial AI voice calling system will identify itself to humans","publisher":"The Verge","type":"news","date":"2018-05-10"}]}
{"id":"2018-06-11-gpt-1","date":"2018-06-11","precision":"day","title":"OpenAI pretrains a transformer on unlabeled text","summary":"The first GPT was small and its results were modest. The recipe was the point: pretrain on whatever text exists, then fine-tune for the task you actually want.","category":"model","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/language-unsupervised/","title":"Improving language understanding with unsupervised learning","publisher":"OpenAI","type":"announcement","date":"2018-06-11"}}
{"id":"2018-10-11-bert","date":"2018-10-11","precision":"day","title":"BERT learns language from both directions","summary":"Google trained BERT to recover hidden words using context on both sides, then adapted it to tasks such as answering questions from a passage. The paper reported leading results on eleven language tasks; code and pretrained models followed later.","category":"research","orgs":["Google"],"tier":2,"source":{"url":"https://arxiv.org/abs/1810.04805","title":"BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding","publisher":"arXiv","type":"paper","date":"2018-10-11"}}
{"id":"2018-12-02-alphafold-casp13","date":"2018-12-02","precision":"day","title":"AlphaFold wins CASP13 on its first entry","summary":"DeepMind entered the protein structure competition for the first time and placed first, ahead of groups who had spent careers on the problem. Structural biologists understood what had happened well before anyone else did.","category":"research","orgs":["DeepMind"],"tier":3,"source":{"url":"https://deepmind.google/discover/blog/alphafold-using-ai-for-scientific-discovery/","title":"AlphaFold: Using AI for scientific discovery","publisher":"Google DeepMind","type":"announcement","date":"2018-12-02","archiveUrl":"https://web.archive.org/web/20181203013114/https://deepmind.com/blog/alphafold/"},"references":[{"url":"https://www.theguardian.com/science/2018/dec/02/google-deepminds-ai-program-alphafold-predicts-3d-shapes-of-proteins","title":"Google's DeepMind predicts 3D shapes of proteins","publisher":"The Guardian","type":"news","date":"2018-12-02"},{"url":"https://moalquraishi.wordpress.com/2018/12/09/alphafold-casp13-what-just-happened/","title":"AlphaFold @ CASP13: \"What just happened?\"","publisher":"Mohammed AlQuraishi","type":"analysis","date":"2018-12-09"}]}
{"id":"2019-02-14-gpt-2-staged-release","date":"2019-02-14","precision":"day","title":"OpenAI withholds GPT-2 and starts an argument","summary":"The lab declined to release the full model on misuse grounds and was accused of both hype and paternalism, often by the same people. The staged-release debate it started is now standard practice across the field.","category":"model","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/better-language-models/","title":"Better language models and their implications","publisher":"OpenAI","type":"announcement","date":"2019-02-14"}}
{"id":"2019-07-22-microsoft-invests-in-openai","date":"2019-07-22","precision":"day","title":"Microsoft invests $1 billion in OpenAI","summary":"A billion dollars, an exclusive Azure partnership, and a capped-profit structure OpenAI had invented four months earlier. The compute-for-equity template every frontier lab now runs on starts here.","category":"funding","orgs":["Microsoft","OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/microsoft-invests-in-and-partners-with-openai/","title":"Microsoft invests in and partners with OpenAI to support us building beneficial AGI","publisher":"OpenAI","type":"announcement","date":"2019-07-22"}}
{"id":"2020-01-23-neural-scaling-laws","date":"2020-01-23","precision":"day","title":"OpenAI's scaling laws paper turns model size into a budgeting decision","summary":"Loss falls as a power law in parameters, data and compute, measured across six orders of magnitude of model size. The paper's advice was to spend a larger budget mostly on a larger model, which is how GPT-3 was sized and what Chinchilla spent 2022 correcting.","category":"research","orgs":["OpenAI"],"tier":2,"source":{"url":"https://arxiv.org/abs/2001.08361","title":"Scaling Laws for Neural Language Models","publisher":"arXiv","type":"paper","date":"2020-01-23"},"references":[{"url":"https://arxiv.org/abs/2203.15556","title":"Training Compute-Optimal Large Language Models","publisher":"arXiv","type":"paper","date":"2022-03-29"},{"url":"https://arxiv.org/abs/2406.19146","title":"Resolving Discrepancies in Compute-Optimal Scaling of Language Models","publisher":"arXiv","type":"paper","date":"2024-06-27"},{"url":"https://arxiv.org/abs/2406.12907","title":"Reconciling Kaplan and Chinchilla Scaling Laws","publisher":"arXiv","type":"paper","date":"2024-06-12"}]}
{"id":"2020-05-28-gpt-3","date":"2020-05-28","precision":"day","title":"GPT-3 learns tasks from the prompt","summary":"OpenAI's 175-billion-parameter GPT-3 performed tasks from instructions and examples in its prompt, without task-specific weight updates in the reported evaluations. The paper tested this approach across language tasks; private-beta API access followed in June.","category":"model","orgs":["OpenAI"],"tier":1,"source":{"url":"https://arxiv.org/abs/2005.14165","title":"Language Models are Few-Shot Learners","publisher":"arXiv","type":"paper","date":"2020-05-28"}}
{"id":"2020-06-11-openai-api","date":"2020-06-11","precision":"day","title":"OpenAI sells access instead of weights","summary":"GPT-3 shipped as a waitlisted HTTP endpoint rather than a download. That single packaging decision defined how the industry would buy intelligence for the next five years.","category":"product","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/openai-api/","title":"OpenAI API","publisher":"OpenAI","type":"announcement","date":"2020-06-11"}}
{"id":"2020-11-30-alphafold-2","date":"2020-11-30","precision":"day","title":"AlphaFold 2 predicts protein structures with high accuracy at CASP14","summary":"DeepMind's AlphaFold 2 produced highly accurate protein-structure predictions in CASP14's blind assessment. It gave researchers a way to infer a protein's shape from its sequence, while interactions and biological function still required further investigation.","category":"research","orgs":["DeepMind"],"tier":1,"source":{"url":"https://deepmind.google/discover/blog/alphafold-a-solution-to-a-50-year-old-grand-challenge-in-biology/","title":"AlphaFold: a solution to a 50-year-old grand challenge in biology","publisher":"Google DeepMind","type":"announcement","date":"2020-11-30"}}
{"id":"2021-01-05-dall-e-and-clip","date":"2021-01-05","precision":"day","title":"OpenAI ships DALL·E and CLIP on the same day","summary":"OpenAI introduced DALL-E, which generated images from text, and CLIP, which learned to associate images with text. CLIP also supported ranking generated images against prompts.","category":"model","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/dall-e/","title":"DALL·E: Creating images from text","publisher":"OpenAI","type":"announcement","date":"2021-01-05"}}
{"id":"2021-05-28-anthropic-founded","date":"2021-05-28","precision":"day","title":"Anthropic launches with $124 million","summary":"Seven people left OpenAI, including its VP of research, to start a lab organized around AI safety research. The split reshaped the competitive structure of the field and gave it a second pole.","category":"company","orgs":["Anthropic"],"tier":1,"source":{"url":"https://techcrunch.com/2021/05/28/anthropic-is-the-new-ai-research-outfit-from-openais-dario-amodei-and-it-has-124m-to-burn/","title":"Anthropic is the new AI research outfit from OpenAI's Dario Amodei, and it has $124M to burn","publisher":"TechCrunch","author":"Devin Coldewey","type":"news","date":"2021-05-28"}}
{"id":"2021-06-17-lora","date":"2021-06-17","precision":"day","title":"LoRA makes fine-tuning a 35MB file instead of a second copy of the model","summary":"Freeze the pretrained weights, train a thin pair of matrices beside them, and a task-specific GPT-3 checkpoint drops from 350GB to 35MB. Microsoft's method became the default way to adapt an open-weight model, and once quantization was stacked on top, the reason one consumer GPU could do it.","category":"research","orgs":["Microsoft"],"tier":2,"source":{"url":"https://arxiv.org/abs/2106.09685","title":"LoRA: Low-Rank Adaptation of Large Language Models","publisher":"arXiv","type":"paper","date":"2021-06-17"},"references":[{"url":"https://arxiv.org/abs/2305.14314","title":"QLoRA: Efficient Finetuning of Quantized LLMs","publisher":"arXiv","type":"paper","date":"2023-05-23"},{"url":"https://huggingface.co/blog/peft","title":"PEFT: Parameter-Efficient Fine-Tuning of Billion-Scale Models on Low-Resource Hardware","publisher":"Hugging Face","type":"docs","date":"2023-02-10"},{"url":"https://arxiv.org/abs/2405.09673","title":"LoRA Learns Less and Forgets Less","publisher":"arXiv","type":"paper","date":"2024-05-15"}]}
{"id":"2021-06-29-github-copilot","date":"2021-06-29","precision":"day","title":"GitHub Copilot writes code in the editor","summary":"An editor assistant that suggests code from comments and surrounding context, released as a technical preview. Generated functions moved into the same window where developers would have to review them.","category":"product","orgs":["GitHub","OpenAI"],"tier":1,"source":{"url":"https://github.blog/news-insights/product-news/introducing-github-copilot-ai-pair-programmer/","title":"Introducing GitHub Copilot: your AI pair programmer","publisher":"GitHub","type":"announcement","date":"2021-06-29"},"references":[{"url":"https://arxiv.org/abs/2302.06590v1","title":"The Impact of AI on Developer Productivity: Evidence from GitHub Copilot","publisher":"arXiv","type":"paper","date":"2023-02-13"}]}
{"id":"2021-08-10-openai-codex","date":"2021-08-10","precision":"day","title":"OpenAI opens Codex to a private beta","summary":"The model underneath Copilot, handed directly to anyone who wanted to build their own version of it. Natural language in, working Python out, and the arguments about training data starting up in the background.","category":"model","orgs":["OpenAI"],"tier":3,"source":{"url":"https://openai.com/index/openai-codex/","title":"OpenAI Codex","publisher":"OpenAI","type":"announcement","date":"2021-08-10","archiveUrl":"https://web.archive.org/web/20240502045910/https://openai.com/index/openai-codex/"},"references":[{"url":"https://techcrunch.com/2021/08/10/openai-upgrades-its-natural-language-ai-coder-codex-and-kicks-off-private-beta/","title":"OpenAI upgrades its natural language AI coder Codex and kicks off private beta","publisher":"TechCrunch","type":"news","date":"2021-08-10"}]}
{"id":"2022-01-27-instructgpt","date":"2022-01-27","precision":"day","title":"InstructGPT becomes the OpenAI API default and RLHF reaches customers","summary":"OpenAI made instruction-tuned models the default on its API, a year after putting an earlier version into beta. Labelers preferred a 1.3B model trained on their own rankings to the 175B GPT-3, and ten months later ChatGPT shipped on the same method.","category":"research","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/instruction-following/","title":"Aligning language models to follow instructions","publisher":"OpenAI","type":"announcement","date":"2022-01-27"},"references":[{"url":"https://arxiv.org/abs/2203.02155","title":"Training language models to follow instructions with human feedback","publisher":"arXiv","type":"paper","date":"2022-03-04"},{"url":"https://www.technologyreview.com/2022/01/27/1044398/new-gpt3-openai-chatbot-language-model-ai-toxic-misinformation/","title":"The new version of GPT-3 is much better behaved (and should be less toxic)","publisher":"MIT Technology Review","type":"news","date":"2022-01-27","author":"Will Douglas Heaven"}]}
{"id":"2022-01-28-chain-of-thought-prompting","date":"2022-01-28","precision":"day","title":"Chain-of-thought prompting gets reasoning out of a model without training it","summary":"Put a few worked examples with their intermediate steps in the prompt and a large enough model starts showing its work, with no fine-tuning and no gradient updates anywhere. Below roughly 100 billion parameters it does nothing, or makes the answers worse.","category":"research","orgs":["Google"],"tier":2,"source":{"url":"https://arxiv.org/abs/2201.11903","title":"Chain-of-Thought Prompting Elicits Reasoning in Large Language Models","publisher":"arXiv","type":"paper","date":"2022-01-28"},"references":[{"url":"https://arxiv.org/abs/2304.15004","title":"Are Emergent Abilities of Large Language Models a Mirage?","publisher":"arXiv","type":"paper","date":"2023-04-28"},{"url":"https://arxiv.org/abs/2305.04388","title":"Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting","publisher":"arXiv","type":"paper","date":"2023-05-07"},{"url":"https://arxiv.org/abs/2409.12183","title":"To CoT or not to CoT? Chain-of-thought helps mainly on math and symbolic reasoning","publisher":"arXiv","type":"paper","date":"2024-09-18"}]}
{"id":"2022-03-29-chinchilla","date":"2022-03-29","precision":"day","title":"Chinchilla finds the big models were undertrained, not undersized","summary":"DeepMind trained more than 400 models to conclude that parameters and tokens should scale together, then showed it with a 70B model on 1.4 trillion tokens that beat the 280B Gopher at equal compute. Two years of frontier training had been spending the budget the wrong way.","category":"research","orgs":["DeepMind"],"tier":2,"source":{"url":"https://arxiv.org/abs/2203.15556","title":"Training Compute-Optimal Large Language Models","publisher":"arXiv","type":"paper","date":"2022-03-29"},"references":[{"url":"https://deepmind.google/blog/an-empirical-analysis-of-compute-optimal-large-language-model-training/","title":"An empirical analysis of compute-optimal large language model training","publisher":"Google DeepMind","type":"announcement","date":"2022-04-12"},{"url":"https://arxiv.org/abs/2404.10102","title":"Chinchilla Scaling: A replication attempt","publisher":"arXiv","type":"paper","date":"2024-04-15"},{"url":"https://arxiv.org/abs/2302.13971","title":"LLaMA: Open and Efficient Foundation Language Models","publisher":"arXiv","type":"paper","date":"2023-02-27"}]}
{"id":"2022-04-06-dall-e-2","date":"2022-04-06","precision":"day","title":"DALL·E 2 makes the waitlist a status symbol","summary":"OpenAI introduced DALL-E 2 with higher-resolution image generation and inpainting, initially limiting access through a waiting list.","category":"model","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/dall-e-2/","title":"DALL·E 2","publisher":"OpenAI","type":"announcement","date":"2022-04-06"}}
{"id":"2022-05-27-flashattention","date":"2022-05-27","precision":"day","title":"FlashAttention speeds up attention by moving less memory, not less math","summary":"Tiling and recomputation keep the large attention matrix out of GPU high bandwidth memory, cutting GPT-2 training time by roughly three times and making memory use linear in sequence length. It computes exactly the same function, which is why it displaced a generation of approximate attention.","category":"research","orgs":["Stanford University"],"tier":2,"source":{"url":"https://arxiv.org/abs/2205.14135","title":"FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness","publisher":"arXiv","type":"paper","date":"2022-05-27"},"references":[{"url":"https://arxiv.org/abs/2307.08691","title":"FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning","publisher":"arXiv","type":"paper","date":"2023-07-17"},{"url":"https://arxiv.org/abs/2407.08608","title":"FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision","publisher":"arXiv","type":"paper","date":"2024-07-11"},{"url":"https://tridao.me/blog/2024/flash3/","title":"FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision","publisher":"Tri Dao","type":"analysis","date":"2024-07-11","author":"Tri Dao"}]}
{"id":"2022-08-22-stable-diffusion-public-release","date":"2022-08-22","precision":"day","title":"Stable Diffusion puts image generation on a home GPU","summary":"Stable Diffusion released downloadable weights for text-to-image generation, letting developers run and adapt the model themselves. It generated images through a compressed representation that reduced compute demands, with responsibilities and distribution requirements under its release license.","category":"model","secondaryCategories":["open-source"],"orgs":["Stability AI","CompVis","Runway"],"tier":1,"source":{"url":"https://stability.ai/news/stable-diffusion-public-release","title":"Stable Diffusion Public Release","publisher":"Stability AI","type":"announcement","date":"2022-08-22"}}
{"id":"2022-11-30-chatgpt-launch","date":"2022-11-30","precision":"day","title":"OpenAI launches ChatGPT","summary":"A chat interface on GPT-3.5, released as a research preview with low expectations attached. It hit a million users in five days and set off everything that followed.","category":"product","orgs":["OpenAI"],"tier":1,"source":{"url":"https://openai.com/index/chatgpt/","title":"Introducing ChatGPT","publisher":"OpenAI","type":"announcement","date":"2022-11-30"},"references":[{"url":"https://www.technologyreview.com/2023/02/08/1068068/chatgpt-is-everywhere-heres-where-it-came-from/","title":"ChatGPT is everywhere. Here's where it came from","publisher":"MIT Technology Review","author":"Will Douglas Heaven","type":"analysis","date":"2023-02-08"}]}
{"id":"2023-01-23-microsoft-extends-openai-partnership","date":"2023-01-23","precision":"day","title":"Microsoft extends the OpenAI partnership by a reported $10 billion","summary":"Microsoft and OpenAI extended their partnership through a multiyear, multibillion-dollar investment, continuing Azure’s role as OpenAI’s exclusive cloud provider. The exact dollar amount was not disclosed.","category":"funding","orgs":["Microsoft","OpenAI"],"tier":2,"source":{"url":"https://blogs.microsoft.com/blog/2023/01/23/microsoftandopenaiextendpartnership/","title":"Microsoft and OpenAI extend partnership","publisher":"Microsoft","type":"announcement","date":"2023-01-23"},"references":[{"url":"https://www.cnbc.com/2023/01/23/microsoft-announces-multibillion-dollar-investment-in-chatgpt-maker-openai.html","title":"Microsoft announces new multibillion-dollar investment in ChatGPT-maker OpenAI","publisher":"CNBC","type":"news","date":"2023-01-23"}]}
{"id":"2023-02-24-llama","date":"2023-02-24","precision":"day","title":"Meta releases LLaMA to researchers, and then to everyone","summary":"Weights for a 65B model, handed out on request and circulating on BitTorrent within a week. The leak created the open-weight ecosystem that Meta then decided to lead on purpose.","category":"model","secondaryCategories":["open-source"],"orgs":["Meta"],"tier":1,"source":{"url":"https://ai.meta.com/blog/large-language-model-llama-meta-ai/","title":"Introducing LLaMA: A foundational, 65-billion-parameter large language model","publisher":"Meta AI","type":"announcement","date":"2023-02-24"}}
{"id":"2023-03-14-claude-launch","date":"2023-03-14","precision":"day","title":"Anthropic introduces Claude","summary":"Anthropic introduced its Claude assistant on March 14, 2023, the same day OpenAI announced GPT-4. The company described its approach as constitutional AI.","category":"model","orgs":["Anthropic"],"tier":1,"source":{"url":"https://www.anthropic.com/news/introducing-claude","title":"Introducing Claude","publisher":"Anthropic","type":"announcement","date":"2023-03-14"}}
{"id":"2023-03-14-gpt-4","date":"2023-03-14","precision":"day","title":"GPT-4 arrives with a technical report that says nothing","summary":"OpenAI released GPT-4 text input through ChatGPT and a waitlisted API; image inputs remained a research preview and were not publicly available at launch. OpenAI reported strong exam results, including on the bar exam. Its technical report withheld architecture, dataset and training-compute details.","category":"model","orgs":["OpenAI"],"tier":1,"source":{"url":"https://openai.com/index/gpt-4-research/","title":"GPT-4","publisher":"OpenAI","type":"announcement","date":"2023-03-14"},"references":[{"url":"https://arxiv.org/abs/2303.08774","title":"GPT-4 Technical Report","publisher":"arXiv","type":"paper","date":"2023-03-15"}]}
{"id":"2023-05-29-direct-preference-optimization","date":"2023-05-29","precision":"day","title":"Direct Preference Optimization takes the reward model out of RLHF","summary":"The objective RLHF reaches with a separate reward model and a reinforcement learning loop turns out to be solvable directly on preference pairs, as something close to a classification problem. The experiments stopped at 6B parameters; Meta later used the method on a 405B one.","category":"research","orgs":["Stanford University"],"tier":2,"source":{"url":"https://arxiv.org/abs/2305.18290","title":"Direct Preference Optimization: Your Language Model is Secretly a Reward Model","publisher":"arXiv","type":"paper","date":"2023-05-29"},"references":[{"url":"https://arxiv.org/abs/2407.21783","title":"The Llama 3 Herd of Models","publisher":"arXiv","type":"paper","date":"2024-07-31"},{"url":"https://arxiv.org/abs/2404.10719","title":"Is DPO Superior to PPO for LLM Alignment? A Comprehensive Study","publisher":"arXiv","type":"paper","date":"2024-04-16"},{"url":"https://blog.neurips.cc/2023/12/11/announcing-the-neurips-2023-paper-awards/","title":"Announcing the NeurIPS 2023 Paper Awards","publisher":"NeurIPS","type":"news","date":"2023-12-11"}]}
{"id":"2023-07-11-claude-2","date":"2023-07-11","precision":"day","title":"Claude 2 opens to the public","summary":"Anthropic made Claude 2 available through a public sign-up, giving more users access to its existing 100,000-token context window.","category":"model","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/claude-2","title":"Claude 2","publisher":"Anthropic","type":"announcement","date":"2023-07-11"}}
{"id":"2023-07-12-xai-founded","date":"2023-07-12","precision":"day","title":"Elon Musk announces xAI","summary":"A twelve person team, a mission statement about understanding the universe, and a launch date picked because its digits add up to 42. The lab shipped a chatbot within four months.","category":"company","orgs":["xAI"],"tier":2,"source":{"url":"https://www.foxbusiness.com/technology/elon-musk-introduces-xai-artificial-intelligence-company","title":"Elon Musk introduces xAI, a new artificial intelligence company","publisher":"Fox Business","type":"news","date":"2023-07-12"}}
{"id":"2023-07-18-llama-2","date":"2023-07-18","precision":"day","title":"Llama 2 ships with a commercial license","summary":"Downloadable models for research and commercial use under a community license, with acceptable-use rules and additional terms for very large services. Access to weights had become a distribution strategy, with conditions attached.","category":"model","secondaryCategories":["open-source"],"orgs":["Meta","Microsoft"],"tier":1,"source":{"url":"https://ai.meta.com/blog/llama-2/","title":"Meta and Microsoft Introduce the Next Generation of Llama","publisher":"Meta AI","type":"announcement","date":"2023-07-18"},"references":[{"url":"https://github.com/meta-llama/llama/blob/llama_v2/LICENSE","title":"LLAMA 2 COMMUNITY LICENSE AGREEMENT","publisher":"Meta","type":"docs","date":"2023-07-18"}]}
{"id":"2023-09-25-amazon-invests-in-anthropic","date":"2023-09-25","precision":"day","title":"Amazon commits up to $4 billion to Anthropic","summary":"An initial $1.25 billion, AWS as the primary cloud provider, and Trainium as the training chip. A hyperscaler funding a lab to rent back its own compute was now the shape of the industry.","category":"funding","orgs":["Amazon","Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/news/anthropic-amazon","title":"Expanding access to safer AI with Amazon","publisher":"Anthropic","type":"announcement","date":"2023-09-25"},"references":[{"url":"https://www.cnbc.com/2023/09/25/amazon-to-invest-up-to-4-billion-in-anthropic-a-rival-to-chatgpt-developer-openai.html","title":"Amazon to invest up to $4 billion in Anthropic, a rival to ChatGPT developer OpenAI","publisher":"CNBC","type":"news","date":"2023-09-25"}]}
{"id":"2023-09-27-mistral-7b","date":"2023-09-27","precision":"day","title":"Mistral 7B beats models twice its size","summary":"An Apache-licensed 7B model from a four-month-old Paris company that outperformed 13B competitors. Europe got a frontier lab, and small models stopped being a consolation prize.","category":"model","secondaryCategories":["open-source"],"orgs":["Mistral AI"],"tier":3,"source":{"url":"https://mistral.ai/news/announcing-mistral-7b","title":"Mistral 7B","publisher":"Mistral AI","type":"announcement","date":"2023-09-27"}}
{"id":"2023-11-03-grok-beta","date":"2023-11-03","precision":"day","title":"xAI puts Grok into beta","summary":"A chatbot with live access to posts on X and a setting for answering rudely. Four months from company announcement to shipped product, which was the actual claim being made.","category":"product","orgs":["xAI"],"tier":3,"source":{"url":"https://x.ai/news/grok","title":"Announcing Grok","publisher":"xAI","type":"announcement","date":"2023-11-03"}}
{"id":"2023-11-06-openai-devday-gpts","date":"2023-11-06","precision":"day","title":"OpenAI's first DevDay introduces GPTs","summary":"GPT-4 Turbo, a 128k context window, custom GPTs and a store to sell them in. Eleven days later the chief executive who announced all of it was fired.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://openai.com/index/new-models-and-developer-products-announced-at-devday/","title":"New models and developer products announced at DevDay","publisher":"OpenAI","type":"announcement","date":"2023-11-06"}}
{"id":"2023-11-17-openai-board-fires-altman","date":"2023-11-17","precision":"day","title":"OpenAI's board fires Sam Altman, briefly","summary":"The board said he had not been consistently candid, roughly 700 of 770 employees threatened to resign, and he was back within five days with a new board. The nonprofit governance experiment ended in public.","category":"company","orgs":["OpenAI"],"tier":1,"source":{"url":"https://openai.com/index/openai-announces-leadership-transition/","title":"OpenAI announces leadership transition","publisher":"OpenAI","type":"announcement","date":"2023-11-17"}}
{"id":"2023-12-01-mamba","date":"2023-12-01","precision":"day","title":"Mamba offers a Transformer alternative that carries fixed-size state","summary":"A selective state space model, linear in sequence length, with no attention and no cache that grows as the context does, reported at four to five times a comparable Transformer's inference throughput. What actually shipped afterwards were hybrids that keep a few attention layers for recall.","category":"research","orgs":["Carnegie Mellon University","Princeton University"],"tier":2,"source":{"url":"https://arxiv.org/abs/2312.00752","title":"Mamba: Linear-Time Sequence Modeling with Selective State Spaces","publisher":"arXiv","type":"paper","date":"2023-12-01"},"references":[{"url":"https://arxiv.org/abs/2405.21060","title":"Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality","publisher":"arXiv","type":"paper","date":"2024-05-31"},{"url":"https://arxiv.org/abs/2402.01032","title":"Repeat After Me: Transformers are Better than State Space Models at Copying","publisher":"arXiv","type":"paper","date":"2024-02-01"},{"url":"https://arxiv.org/abs/2403.19887","title":"Jamba: A Hybrid Transformer-Mamba Language Model","publisher":"arXiv","type":"paper","date":"2024-03-28"}]}
{"id":"2023-12-06-gemini","date":"2023-12-06","precision":"day","title":"Google answers with Gemini","summary":"Three model sizes, native multimodality, and a launch video that turned out to have been edited for pace. The model was real and the credibility cost was real too.","category":"model","orgs":["Google"],"tier":2,"source":{"url":"https://blog.google/technology/ai/google-gemini-ai/","title":"Introducing Gemini: our largest and most capable AI model","publisher":"Google","type":"announcement","date":"2023-12-06"}}
{"id":"2024-02-15-sora","date":"2024-02-15","precision":"day","title":"Sora generates a minute of coherent video","summary":"OpenAI demonstrated Sora generating videos up to a minute long and initially limited access to red teamers and selected creative professionals. A public product followed in December.","category":"model","orgs":["OpenAI"],"tier":3,"source":{"url":"https://openai.com/index/sora/","title":"Sora: Creating video from text","publisher":"OpenAI","type":"announcement","date":"2024-02-15"}}
{"id":"2024-03-04-claude-3","date":"2024-03-04","precision":"day","title":"Claude 3 takes the top of the benchmarks","summary":"Anthropic introduced the Claude 3 family: Haiku, Sonnet and Opus. It reported that Opus surpassed GPT-4 on several benchmark tests at launch.","category":"model","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/news/claude-3-family","title":"Introducing the next generation of Claude","publisher":"Anthropic","type":"announcement","date":"2024-03-04"}}
{"id":"2024-03-13-eu-ai-act-adopted","date":"2024-03-13","precision":"day","title":"The European Parliament adopts the AI Act","summary":"The European Parliament adopted the AI Act, which sets requirements according to system risk and applies to some providers outside the EU. Its provisions take effect in stages.","category":"policy","orgs":["European Union"],"tier":2,"source":{"url":"https://www.europarl.europa.eu/news/en/press-room/20240308IPR19015/artificial-intelligence-act-meps-adopt-landmark-law","title":"Artificial Intelligence Act: MEPs adopt landmark law","publisher":"European Parliament","type":"press-release","date":"2024-03-13"}}
{"id":"2024-04-16-fhi-closes","date":"2024-04-16","precision":"day","title":"Oxford's Future of Humanity Institute closes after 19 years","summary":"The Faculty of Philosophy declined to renew the remaining contracts and the institute that coined much of the existential-risk vocabulary, and housed Bostrom while he wrote Superintelligence, shut its doors with no announcement. Its final report was an oral history of itself.","category":"company","orgs":["Future of Humanity Institute"],"tier":3,"source":{"url":"https://www.futureofhumanityinstitute.org/","title":"Future of Humanity Institute","publisher":"Future of Humanity Institute","type":"docs","retrieved":"2026-09-06","locator":"On 16 April 2024, the Institute was closed down."}}
{"id":"2024-04-18-llama-3","date":"2024-04-18","precision":"day","title":"Llama 3 narrows the gap to the closed models","summary":"Meta shipped 8B and 70B weights that held their own against paid frontier models on most public evaluations. Open weights went from cheap alternative to genuine option.","category":"model","secondaryCategories":["open-source"],"orgs":["Meta"],"tier":3,"source":{"url":"https://ai.meta.com/blog/meta-llama-3/","title":"Introducing Meta Llama 3: The most capable openly available LLM to date","publisher":"Meta AI","type":"announcement","date":"2024-04-18"}}
{"id":"2024-04-30-alden-newspapers-sue-openai","date":"2024-04-30","precision":"day","title":"Eight Alden Global Capital newspapers sue OpenAI and Microsoft","summary":"The Chicago Tribune, New York Daily News, Denver Post and five sister papers filed in the Southern District of New York, the same court as the Times case, alleging millions of articles were copied to train ChatGPT and Copilot. They asked for damages, an injunction and the destruction of any model built on their work.","category":"policy","orgs":["Alden Global Capital","OpenAI","Microsoft"],"tier":3,"source":{"url":"https://chicago.suntimes.com/money/2024/04/30/chicago-tribune-newspapers-media-news-group-alden-global-microsoft-openai","title":"Chicago Tribune, seven other newspapers sue Microsoft and OpenAI","publisher":"Chicago Sun-Times","type":"news","date":"2024-04-30"},"references":[{"url":"https://www.axios.com/2024/04/30/microsoft-openai-lawsuit-copyright-newspapers-alden-global","title":"Major U.S. newspapers sue OpenAI, Microsoft for copyright infringement","publisher":"Axios","type":"news","date":"2024-04-30"}]}
{"id":"2024-05-01-claude-team-plan-ios","date":"2024-05-01","precision":"day","title":"Anthropic launches the Claude Team plan and an iOS app","summary":"Thirty dollars a seat, five seats minimum, with the 200K context window as the pitch to businesses. The iOS app arrived a year after ChatGPT's, and the announcement listed the business plan first.","category":"product","orgs":["Anthropic"],"tier":3,"source":{"url":"https://claude.com/blog/team-plan-and-ios","title":"Introducing the Claude Team plan and iOS app","publisher":"Anthropic","type":"announcement","date":"2024-05-01"}}
{"id":"2024-05-06-stack-overflow-openai-partnership","date":"2024-05-06","precision":"day","title":"Stack Overflow licenses its answers to OpenAI","summary":"OpenAI gets the OverflowAPI feed and shows attributed Stack Overflow answers inside ChatGPT; Stack Overflow gets OpenAI models for OverflowAI. Users who edited their answers in protest had them restored by moderators and their accounts suspended.","category":"deal","orgs":["Stack Overflow","OpenAI"],"tier":3,"source":{"url":"https://stackoverflow.co/company/press/archive/openai-partnership","title":"Stack Overflow and OpenAI Partner to Strengthen the World's Most Popular Large Language Models","publisher":"Stack Overflow","type":"press-release","date":"2024-05-06"}}
{"id":"2024-05-07-huawei-export-licenses-revoked","date":"2024-05-07","precision":"day","title":"US revokes Intel and Qualcomm licenses to sell chips to Huawei","summary":"The Commerce Department cancelled the licenses that let Intel and Qualcomm ship laptop and handset chips to Huawei, with immediate effect. Weeks earlier, Huawei had launched an AI laptop running on an Intel chip, and Congress wanted to know why that was possible.","category":"policy","orgs":["United States Department of Commerce","Intel","Qualcomm","Huawei"],"tier":3,"source":{"url":"https://www.trendforce.com/news/2024/05/08/news-u-s-imposes-further-sanctions-revoking-intel-and-qualcomms-license-to-supply-chips-to-huawei/","title":"U.S. Imposes Further Sanctions, Revoking Intel and Qualcomm's License to Supply Chips to Huawei","publisher":"TrendForce","type":"news","date":"2024-05-08"}}
{"id":"2024-05-08-microsoft-wisconsin-datacenter","date":"2024-05-08","precision":"day","title":"Microsoft commits $3.3 billion to a Wisconsin AI datacenter","summary":"A Mount Pleasant campus on the site Foxconn once promised to fill, announced with President Biden on stage, plus a manufacturing AI lab at UW-Milwaukee and a pledge to train 100,000 residents. One of the first datacenter announcements framed as regional economic policy rather than cloud capacity.","category":"hardware","orgs":["Microsoft"],"tier":3,"source":{"url":"https://news.microsoft.com/source/2024/05/08/microsoft-announces-3-3-billion-investment-in-wisconsin-to-spur-artificial-intelligence-innovation-and-economic-growth/","title":"Microsoft announces $3.3 billion investment in Wisconsin to spur artificial intelligence innovation and economic growth","publisher":"Microsoft","type":"announcement","date":"2024-05-08"}}
{"id":"2024-05-08-openai-model-spec","date":"2024-05-08","precision":"day","title":"OpenAI publishes the first Model Spec","summary":"A public document saying how OpenAI wants its models to behave, split into objectives, rules and defaults, and written to be used as RLHF guidance. Headlines fixed on one line about exploring NSFW content in age-appropriate contexts.","category":"research","orgs":["OpenAI"],"tier":3,"source":{"url":"https://openai.com/index/introducing-the-model-spec/","title":"Introducing the Model Spec","publisher":"OpenAI","type":"announcement","date":"2024-05-08"},"references":[{"url":"https://cdn.openai.com/spec/model-spec-2024-05-08.html","title":"Model Spec (2024/05/08)","publisher":"OpenAI","type":"docs","date":"2024-05-08"}]}
{"id":"2024-05-13-gpt-4o","date":"2024-05-13","precision":"day","title":"GPT-4o talks back in real time","summary":"One model handling text, audio and vision natively, with conversational latency and a free tier. The demo was charming and the pricing was the actual weapon.","category":"model","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/hello-gpt-4o/","title":"Hello GPT-4o","publisher":"OpenAI","type":"announcement","date":"2024-05-13"}}
{"id":"2024-05-14-gemini-1-5-flash","date":"2024-05-14","precision":"day","title":"Gemini 1.5 Flash and Project Astra debut at Google I/O","summary":"Google's answer to GPT-4o's launch day: Gemini 1.5 Flash for cheap high-volume work, a two million token context window, and Project Astra, a camera-and-voice assistant demo. Flash became the workhorse; Astra stayed a demo.","category":"model","orgs":["Google","Google DeepMind"],"tier":3,"source":{"url":"https://blog.google/technology/ai/google-gemini-update-flash-ai-assistant-io-2024/","title":"Gemini breaks new ground with a faster model, longer context, AI agents and more","publisher":"Google","type":"announcement","date":"2024-05-14"}}
{"id":"2024-05-14-google-ai-overviews","date":"2024-05-14","precision":"day","title":"Google puts AI Overviews on Search in the US","summary":"A generated answer above the links, shipped into the product that funds the company. Publishers watched their referral traffic and started doing arithmetic.","category":"product","orgs":["Google"],"tier":2,"source":{"url":"https://blog.google/products/search/generative-ai-google-search-may-2024/","title":"Generative AI in Search: Let Google do the searching for you","publisher":"Google","type":"announcement","date":"2024-05-14"}}
{"id":"2024-05-14-sutskever-leaves-openai","date":"2024-05-14","precision":"day","title":"Ilya Sutskever leaves OpenAI and the superalignment team dissolves","summary":"The chief scientist who voted to remove Sam Altman departs six months after the reversal, and superalignment co-lead Jan Leike follows within days, writing that safety culture had 'taken a backseat to shiny products'. Their team is folded into the broader research organization.","category":"company","orgs":["OpenAI"],"tier":2,"source":{"url":"https://time.com/6978195/ilya-sutskever-leaves-open-ai/","title":"Ilya Sutskever, Co-Founder and Chief Scientist, Leaves OpenAI","publisher":"TIME","type":"news","date":"2024-05-15"},"references":[{"url":"https://www.cnbc.com/2024/05/17/openai-superalignment-sutskever-leike.html","title":"OpenAI dissolves team focused on long-term AI risks, less than one year after announcing it","publisher":"CNBC","type":"news","date":"2024-05-17"}]}
{"id":"2024-05-16-openai-reddit-deal","date":"2024-05-16","precision":"day","title":"OpenAI signs Reddit for data and distribution","summary":"Reddit's data API feeds OpenAI's models and ChatGPT gets Reddit content, while OpenAI becomes a Reddit advertising partner. Twenty years of forum arguments, now training data with a price tag.","category":"deal","orgs":["OpenAI","Reddit"],"tier":3,"source":{"url":"https://openai.com/index/openai-and-reddit-partnership/","title":"OpenAI and Reddit Partnership","publisher":"OpenAI","type":"announcement","date":"2024-05-16"}}
{"id":"2024-05-16-sony-music-ai-opt-out","date":"2024-05-16","precision":"day","title":"Sony Music tells 700 AI companies not to train on its catalog","summary":"A 'Declaration of AI Training Opt Out' sent to more than 700 AI developers and streaming services, prohibiting training on Sony recordings, lyrics and artwork and demanding an accounting of any past use. The bluntest move yet by a major label.","category":"culture","orgs":["Sony Music Group"],"tier":3,"source":{"url":"https://techcrunch.com/2024/05/16/sony-music-warns-tech-companies-over-unauthorized-use-of-its-content-to-train-ai/","title":"Sony Music warns tech companies over 'unauthorized' use of its content to train AI","publisher":"TechCrunch","type":"news","date":"2024-05-16"}}
{"id":"2024-05-17-colorado-ai-act","date":"2024-05-17","precision":"day","title":"Colorado enacts the first comprehensive US state AI law","summary":"SB24-205 requires developers and deployers of high-risk AI systems to use reasonable care against algorithmic discrimination, enforced by the attorney general starting in 2026. The first comprehensive state AI law in the United States.","category":"policy","orgs":["State of Colorado"],"tier":3,"source":{"url":"https://leg.colorado.gov/bills/sb24-205","title":"SB24-205: Consumer Protections for Artificial Intelligence","publisher":"Colorado General Assembly","type":"docs","locator":"Bill history: 05/17/2024, Governor Signed","retrieved":"2026-08-20"}}
{"id":"2024-05-19-sky-voice-paused","date":"2024-05-19","precision":"day","title":"OpenAI pulls ChatGPT's Sky voice","summary":"After a week of comparisons to Scarlett Johansson's character in Her, OpenAI paused ChatGPT's Sky voice and explained how its voices were cast. Johansson then said she had declined the role twice and hired counsel, which is not how a product launch is supposed to go.","category":"culture","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/how-the-voices-for-chatgpt-were-chosen/","title":"How the voices for ChatGPT were chosen","publisher":"OpenAI","type":"announcement","date":"2024-05-19"}}
{"id":"2024-05-20-copilot-plus-pcs","date":"2024-05-20","precision":"day","title":"Microsoft introduces Copilot+ PCs, with Recall","summary":"A new class of Windows machines built around a 40 TOPS neural processor requirement, Snapdragon chips first. The flagship feature, Recall, photographed everything on screen into a searchable memory and drew a privacy backlash that delayed it for months.","category":"product","orgs":["Microsoft"],"tier":3,"source":{"url":"https://blogs.microsoft.com/blog/2024/05/20/introducing-copilot-pcs/","title":"Introducing Copilot+ PCs","publisher":"Microsoft","type":"announcement","date":"2024-05-20"}}
{"id":"2024-05-21-scale-series-f","date":"2024-05-21","precision":"day","title":"Scale AI raises $1 billion for the data foundry","summary":"A $1 billion Series F at a $13.8 billion valuation, led by Accel with Amazon and Meta joining. The wager: frontier labs will keep paying for the labeled data their models are trained on, whoever wins.","category":"funding","orgs":["Scale AI"],"tier":3,"source":{"url":"https://scale.com/blog/scale-ai-series-f","title":"Scale's Series F: Expanding the Data Foundry for AI","publisher":"Scale AI","type":"announcement","date":"2024-05-21"}}
{"id":"2024-05-21-scaling-monosemanticity","date":"2024-05-21","precision":"day","title":"Anthropic maps millions of features inside a frontier model","summary":"Scaling Monosemanticity extracted millions of interpretable features from Claude 3 Sonnet, the first detailed map of a production model's internals. Golden Gate Claude, a version steered to obsess over one bridge, made interpretability briefly famous.","category":"research","orgs":["Anthropic"],"tier":3,"source":{"url":"https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html","title":"Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet","publisher":"Anthropic","type":"paper","date":"2024-05-21"}}
{"id":"2024-05-21-seoul-frontier-commitments","date":"2024-05-21","precision":"day","title":"Frontier labs sign the Seoul safety commitments","summary":"At the UK and South Korea's Seoul summit, sixteen frontier labs including OpenAI, Anthropic, Google, Meta and Zhipu AI committed to publish safety frameworks with risk thresholds and to hold back models that breach them. Voluntary, and the signatory list kept growing afterwards.","category":"policy","orgs":["OpenAI","Anthropic","Google","Meta","Microsoft","Amazon","Mistral AI","xAI","Cohere","G42","IBM","Inflection AI","Naver","Samsung Electronics","Technology Innovation Institute","Zhipu AI"],"tier":3,"source":{"url":"https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024","title":"Frontier AI Safety Commitments, AI Seoul Summit 2024","publisher":"Department for Science, Innovation and Technology","type":"docs","date":"2024-05-21"}}
{"id":"2024-05-21-suno-series-b","date":"2024-05-21","precision":"day","title":"Suno raises $125 million for text-to-song","summary":"Lightspeed led, with Nat Friedman, Daniel Gross, Matrix and Founder Collective, eight months after the first product and with 10 million people having made a song. The largest round in music AI to that point, and a year before the major labels sued.","category":"funding","orgs":["Suno"],"tier":3,"source":{"url":"https://suno.com/blog/fundraising-announcement-may-2024","title":"Suno has raised $125 million","publisher":"Suno","type":"announcement","date":"2024-05-21"}}
{"id":"2024-05-22-nvidia-q1-fy2025","date":"2024-05-22","precision":"day","title":"NVIDIA reports revenue up 262 percent and splits its stock ten for one","summary":"A $26 billion quarter, $22.6 billion of it datacenter, and net income of $14.9 billion. Huang used the call to say the company was now on a one-year architecture cadence, which is the line the rest of the industry had to plan around.","category":"culture","orgs":["NVIDIA"],"tier":3,"source":{"url":"https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-first-quarter-fiscal-2025","title":"NVIDIA Announces Financial Results for First Quarter Fiscal 2025","publisher":"NVIDIA","type":"press-release","date":"2024-05-22"}}
{"id":"2024-05-22-openai-news-corp-deal","date":"2024-05-22","precision":"day","title":"OpenAI signs News Corp, the biggest publisher deal yet","summary":"Wall Street Journal, Times of London and New York Post content flows into OpenAI products under a multi-year agreement reported at more than $250 million over five years. Signed while the New York Times was suing instead.","category":"deal","orgs":["OpenAI","News Corp"],"tier":3,"source":{"url":"https://openai.com/index/news-corp-and-openai-sign-landmark-multi-year-global-partnership/","title":"A landmark multi-year global partnership with News Corp","publisher":"OpenAI","type":"announcement","date":"2024-05-22"}}
{"id":"2024-05-26-xai-series-b","date":"2024-05-26","precision":"day","title":"xAI raises $6 billion","summary":"A $6 billion Series B at a reported $24 billion valuation, less than a year after the company was founded, to buy the compute for Grok to chase a two-year head start.","category":"funding","orgs":["xAI"],"tier":3,"source":{"url":"https://x.ai/news/series-b","title":"Series B funding round","publisher":"xAI","type":"announcement","date":"2024-05-26"}}
{"id":"2024-05-28-openai-safety-committee","date":"2024-05-28","precision":"day","title":"OpenAI forms a safety committee as its next frontier model trains","summary":"A board committee led by Bret Taylor and Sam Altman gets 90 days to evaluate OpenAI's safety processes, announced in the same post that confirmed training of the next frontier model had begun. The month's departures made the subtext legible.","category":"company","orgs":["OpenAI"],"tier":3,"source":{"url":"https://openai.com/index/openai-board-forms-safety-and-security-committee/","title":"OpenAI Board Forms Safety and Security Committee","publisher":"OpenAI","type":"announcement","date":"2024-05-28"}}
{"id":"2024-05-29-codestral","date":"2024-05-29","precision":"day","title":"Mistral releases Codestral under a non-production license","summary":"A 22 billion parameter code model covering 80 languages, downloadable but not for commercial use without a separate deal. The new Mistral AI Non-Production License was the news: the company that made its name on Apache 2.0 weights had found the line.","category":"model","orgs":["Mistral AI"],"tier":3,"source":{"url":"https://mistral.ai/news/codestral","title":"Codestral","publisher":"Mistral AI","type":"announcement","date":"2024-05-29"}}
{"id":"2024-05-29-palantir-maven-contract","date":"2024-05-29","precision":"day","title":"The Army gives Palantir $480 million to take Maven to production","summary":"A five-year firm-fixed-price contract for the Maven Smart System, the targeting and intelligence platform that began as a 2021 prototype, expanding it to thousands of users across five combatant commands. Palantir was the only bidder. The ceiling was raised by $795 million a year later.","category":"deal","orgs":["Palantir","United States Army"],"tier":3,"source":{"url":"https://defensescoop.com/2024/05/29/palantir-480-million-army-contract-maven-smart-system-artificial-intelligence/","title":"Palantir lands $480M Army contract for Maven artificial intelligence tech","publisher":"DefenseScoop","type":"news","date":"2024-05-29"},"references":[{"url":"https://breakingdefense.com/2024/05/new-contract-expands-maven-ais-users-from-hundreds-to-thousands-worldwide-palantir-says/","title":"New contract expands Maven AI's users 'from hundreds to thousands' worldwide, Palantir says","publisher":"Breaking Defense","type":"news","date":"2024-05"}]}
{"id":"2024-05-30-openai-influence-ops-report","date":"2024-05-30","precision":"day","title":"OpenAI reports covert influence operations using its models","summary":"OpenAI reported taking down five covert influence operations from Russia, China, Iran and Israel that used its models to write propaganda, and noted none achieved significant reach. The first report of its kind from a frontier lab.","category":"culture","orgs":["OpenAI"],"tier":3,"source":{"url":"https://openai.com/index/disrupting-deceptive-uses-of-AI-by-covert-influence-operations/","title":"Disrupting deceptive uses of AI by covert influence operations","publisher":"OpenAI","type":"announcement","date":"2024-05-30"}}
{"id":"2024-06-10-apple-intelligence","date":"2024-06-10","precision":"day","title":"Apple announces Apple Intelligence, with ChatGPT inside","summary":"The company that builds everything in house put a competitor's model behind Siri for the questions it could not answer. The arrangement held for about eighteen months.","category":"product","orgs":["Apple","OpenAI"],"tier":2,"source":{"url":"https://www.apple.com/newsroom/2024/06/introducing-apple-intelligence-for-iphone-ipad-and-mac/","title":"Introducing Apple Intelligence for iPhone, iPad, and Mac","publisher":"Apple","type":"press-release","date":"2024-06-10"}}
{"id":"2024-06-20-claude-3-5-sonnet","date":"2024-06-20","precision":"day","title":"Claude 3.5 Sonnet beats the flagship at a fifth of the price","summary":"Anthropic reported Claude 3.5 Sonnet ahead of Claude 3 Opus on several benchmarks at one-fifth its token price. It also introduced Artifacts, a side panel for working with generated content.","category":"model","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/news/claude-3-5-sonnet","title":"Introducing Claude 3.5 Sonnet","publisher":"Anthropic","type":"announcement","date":"2024-06-20"}}
{"id":"2024-08-01-eu-ai-act-in-force","date":"2024-08-01","precision":"day","title":"The EU AI Act enters into force","summary":"Adopted in March, in force in August, binding in stages out to 2027. The first general AI law with real penalties attached now existed and had a clock running against it.","category":"policy","orgs":["European Union"],"tier":3,"source":{"url":"https://commission.europa.eu/news/ai-act-enters-force-2024-08-01_en","title":"AI Act enters into force","publisher":"European Commission","type":"press-release","date":"2024-08-01"}}
{"id":"2024-09-12-openai-o1-preview","date":"2024-09-12","precision":"day","title":"o1 trades inference time for reasoning","summary":"A model trained to think before answering, spending seconds of compute per response for large gains on math and code. After four years of scaling the same axis, the field found a second one.","category":"model","orgs":["OpenAI"],"tier":1,"source":{"url":"https://openai.com/index/introducing-openai-o1-preview/","title":"Introducing OpenAI o1-preview","publisher":"OpenAI","type":"announcement","date":"2024-09-12"}}
{"id":"2024-10-01-openai-devday-realtime-api","date":"2024-10-01","precision":"day","title":"OpenAI DevDay debuts a Realtime API for voice apps","summary":"A public beta let developers build speech to speech apps with six voices and low latency, alongside prompt caching, vision fine tuning and model distillation. No new model shipped, an unusual DevDay without one.","category":"product","orgs":["OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2024/10/01/openais-devday-brings-realtime-api-and-other-treats-for-ai-app-developers/","title":"OpenAI's DevDay brings Realtime API and other treats for AI app developers","publisher":"TechCrunch","type":"news","date":"2024-10-01"}}
{"id":"2024-10-02-openai-157-billion-funding","date":"2024-10-02","precision":"day","title":"OpenAI raises $6.6 billion at a $157 billion valuation","summary":"Thrive Capital led the round, with Microsoft, Nvidia, SoftBank and others joining. It was the largest venture round on record and nearly doubled OpenAI's valuation from a year earlier.","category":"funding","orgs":["OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2024/10/02/openai-raises-6-6b-and-is-now-valued-at-157b/","title":"OpenAI raises $6.6B and is now valued at $157B","publisher":"TechCrunch","type":"news","date":"2024-10-02"},"references":[{"url":"https://openai.com/index/scale-the-benefits-of-ai/","title":"New funding to scale the benefits of AI","publisher":"OpenAI","type":"announcement","date":"2024-10-02"}]}
{"id":"2024-10-02-poolside-series-b","date":"2024-10-02","precision":"day","title":"AI coding startup Poolside raises $500 million Series B","summary":"Bain Capital Ventures led the round at a $3 billion valuation, with eBay and Nvidia among the backers. Poolside builds open weight foundation models for software development and said the money would fund 10,000 Nvidia GPUs for training.","category":"funding","orgs":["Poolside"],"tier":3,"source":{"url":"https://techcrunch.com/2024/10/02/ai-coding-startup-poolside-raises-500m-from-ebay-nvidia-and-others/","title":"AI coding startup Poolside raises $500M from eBay, Nvidia, and others","publisher":"TechCrunch","type":"news","date":"2024-10-02"}}
{"id":"2024-10-03-chatgpt-canvas","date":"2024-10-03","precision":"day","title":"ChatGPT gets Canvas, a workspace for writing and code","summary":"A side by side editing surface let users draft documents or code next to the chat, with inline edit requests, a length slider and a one click bug review for code. It rolled out in beta to ChatGPT Plus and Team users.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://techcrunch.com/2024/10/03/openai-launches-new-canvas-chatgpt-interface-tailored-to-writing-and-coding-projects/","title":"OpenAI launches new 'Canvas' ChatGPT interface tailored to writing and coding projects","publisher":"TechCrunch","type":"news","date":"2024-10-03"}}
{"id":"2024-10-04-meta-movie-gen","date":"2024-10-04","precision":"day","title":"Meta previews Movie Gen, a video and audio generation model","summary":"A 30 billion parameter model turns text into 16 second videos, personalizes them from a photo, edits footage with plain language commands and generates matching audio. Meta called it research rather than a product and said it would work with filmmakers before any release.","category":"research","orgs":["Meta"],"tier":3,"source":{"url":"https://ai.meta.com/blog/movie-gen-media-foundation-models-generative-ai-video/","title":"How Meta Movie Gen could usher in a new AI-enabled era for content creators","publisher":"Meta AI","type":"announcement","date":"2024-10-04"}}
{"id":"2024-10-08-nobel-physics-neural-networks","date":"2024-10-08","precision":"day","title":"Hopfield and Hinton win the Nobel Prize in Physics","summary":"John Hopfield and Geoffrey Hinton were honored for foundational discoveries that enable machine learning with artificial neural networks, using tools borrowed from statistical physics that underpin deep learning today.","category":"research","orgs":["Princeton University","University of Toronto"],"tier":2,"source":{"url":"https://www.nobelprize.org/prizes/physics/2024/press-release/","title":"Press release: The Nobel Prize in Physics 2024","publisher":"The Nobel Prize","type":"press-release","date":"2024-10-08"},"references":[{"url":"https://www.nobelprize.org/prizes/physics/2024/popular-information/","title":"The Nobel Prize in Physics 2024 - Popular science background","publisher":"The Nobel Prize","type":"analysis","date":"2024-10-08"}]}
{"id":"2024-10-09-nobel-chemistry-alphafold","date":"2024-10-09","precision":"day","title":"Hassabis and Jumper win the Nobel Prize in Chemistry for AlphaFold","summary":"Half the prize went to Demis Hassabis and John Jumper of Google DeepMind for AlphaFold, which predicts a protein's three dimensional structure from its amino acid sequence. The other half went to David Baker for computational protein design.","category":"research","orgs":["Google DeepMind"],"tier":2,"source":{"url":"https://www.nobelprize.org/prizes/chemistry/2024/press-release/","title":"Press release: The Nobel Prize in Chemistry 2024","publisher":"The Nobel Prize","type":"press-release","date":"2024-10-09"},"references":[{"url":"https://www.nobelprize.org/prizes/chemistry/2024/popular-information/","title":"The Nobel Prize in Chemistry 2024 - Popular information","publisher":"The Nobel Prize","type":"analysis","date":"2024-10-09"}]}
{"id":"2024-10-10-tesla-we-robot-cybercab","date":"2024-10-10","precision":"day","title":"Tesla unveils Cybercab, Robovan and dancing Optimus robots","summary":"At a Los Angeles event called We, Robot, Musk showed 20 driverless Cybercab prototypes he priced under $30,000, a Robovan that can carry up to 20 passengers, and about a dozen Optimus humanoid robots mixing drinks and chatting with guests. Cybercab production is targeted for 2026 or later.","category":"product","orgs":["Tesla"],"tier":2,"source":{"url":"https://techcrunch.com/2024/10/10/tesla-reveals-20-cybercabs-at-we-robot-event/","title":"Tesla reveals 20 Cybercabs at We, Robot event, says you'll be able to buy one for less than $30,000","publisher":"TechCrunch","type":"news","date":"2024-10-10"}}
{"id":"2024-10-14-adobe-firefly-video-model","date":"2024-10-14","precision":"day","title":"Adobe launches the Firefly Video Model in beta at MAX","summary":"Announced as the first publicly available video model built for commercial use, it turns text or a still image into video and extends clips inside Premiere Pro. Adobe said Firefly had generated more than 13 billion images since its 2023 launch.","category":"product","orgs":["Adobe"],"tier":3,"source":{"url":"https://news.adobe.com/news/2024/10/101424-adobe-launches-firefly-video-model","title":"Adobe Launches Firefly Video Model and Enhances Image, Vector and Design Models","publisher":"Adobe","type":"press-release","date":"2024-10-14"}}
{"id":"2024-10-21-ibm-granite-3","date":"2024-10-21","precision":"day","title":"IBM releases Granite 3.0 open language models","summary":"The family spans 2B and 8B general purpose models plus safety focused Guardian and mixture of experts variants, all under an Apache 2.0 license. IBM said pairing the smaller models with enterprise data could match larger frontier models at a fraction of the cost.","category":"model","secondaryCategories":["open-source"],"orgs":["IBM"],"tier":3,"source":{"url":"https://newsroom.ibm.com/2024-10-21-ibm-introduces-granite-3-0-high-performing-ai-models-built-for-business","title":"IBM Introduces Granite 3.0: High Performing AI Models Built for Business","publisher":"IBM","type":"press-release","date":"2024-10-21"}}
{"id":"2024-10-21-perplexity-sued-dow-jones-nyp","date":"2024-10-21","precision":"day","title":"Dow Jones and the New York Post sue Perplexity for copyright","summary":"News Corp's publishers accused the AI search startup of using their journalism to train models and answer queries without a license, seeking damages and an injunction. It was the first major publisher lawsuit against Perplexity.","category":"policy","orgs":["Perplexity","News Corp"],"tier":3,"source":{"url":"https://www.courthousenews.com/news-corp-outlets-hit-bezos-backed-ai-startup-with-copyright-suit/","title":"News Corp outlets hit Bezos-backed AI startup with copyright suit","publisher":"Courthouse News Service","type":"news","date":"2024-10-21"},"references":[{"url":"https://www.theguardian.com/technology/2024/oct/21/rupert-murdoch-ai-lawsuit-new-york-post-dow-jones","title":"Rupert Murdoch's Dow Jones and New York Post sue AI firm for using articles","publisher":"The Guardian","type":"news","date":"2024-10-21"}]}
{"id":"2024-10-22-character-ai-sewell-setzer-lawsuit","date":"2024-10-22","precision":"day","title":"Mother sues Character.AI over teen son's death by suicide","summary":"Megan Garcia alleged that Character.AI's product design, including sexualized conversations with her 14 year old son Sewell Setzer, contributed to his death, and named Google and Alphabet as co-defendants. It became a rallying point for concerns over AI companion safety.","category":"culture","orgs":["Character.AI","Google"],"tier":3,"source":{"url":"https://www.nbcnews.com/tech/characterai-lawsuit-florida-teen-death-rcna176791","title":"Lawsuit claims Character.AI is responsible for teen's suicide","publisher":"NBC News","type":"news","date":"2024-10-23"},"references":[{"url":"https://apnews.com/article/chatbot-ai-lawsuit-suicide-teen-artificial-intelligence-9d48adc572100822fdbc3c90d1456bd0","title":"AI chatbot pushed teen to kill himself, lawsuit alleges","publisher":"AP News","type":"news","date":"2024-10-24"}]}
{"id":"2024-10-22-claude-3-5-sonnet-refresh","date":"2024-10-22","precision":"day","title":"Anthropic releases an upgraded Claude 3.5 Sonnet","summary":"Anthropic made an upgraded Claude 3.5 Sonnet available on October 22, reporting improvements in coding and introducing computer use in public beta.","category":"model","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/3-5-models-and-computer-use","title":"Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku","publisher":"Anthropic","type":"announcement","date":"2024-10-22"}}
{"id":"2024-10-22-claude-computer-use","date":"2024-10-22","precision":"day","title":"Claude gets a mouse and keyboard","summary":"An API that let a model look at a screen, move the cursor and type. It was slow, it made mistakes, and it was obviously the direction of travel.","category":"product","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/3-5-models-and-computer-use","title":"Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku","publisher":"Anthropic","type":"announcement","date":"2024-10-22"}}
{"id":"2024-10-22-runway-act-one","date":"2024-10-22","precision":"day","title":"Runway introduces Act-One for AI character performances","summary":"A tool inside Gen-3 Alpha transfers an actor's facial performance, filmed on an ordinary camera, onto an animated or realistic character. It let one performer voice and animate multiple characters in a scene without motion capture equipment.","category":"product","orgs":["Runway"],"tier":3,"source":{"url":"https://runway.com/research/introducing-act-one","title":"Introducing Act-One","publisher":"Runway","type":"announcement","date":"2024-10-22"}}
{"id":"2024-10-22-stable-diffusion-3-5","date":"2024-10-22","precision":"day","title":"Stability AI releases Stable Diffusion 3.5","summary":"Large and Large Turbo were released on October 22 under the Stability AI Community License, free for non-commercial use and commercial use below $1 million in annual revenue. The Medium variant followed on October 29.","category":"model","secondaryCategories":["open-source"],"orgs":["Stability AI"],"tier":3,"source":{"url":"https://stability.ai/news-updates/introducing-stable-diffusion-3-5","title":"Introducing Stable Diffusion 3.5","publisher":"Stability AI","type":"announcement","date":"2024-10-22"}}
{"id":"2024-10-24-cerebras-inference-3x-faster","date":"2024-10-24","precision":"day","title":"Cerebras triples its inference speed to 2,100 tokens per second","summary":"Running Llama 3.1 70B, Cerebras said its wafer scale inference platform now beats the fastest GPU setup by 16 times, fast enough for real time voice and reasoning applications.","category":"hardware","orgs":["Cerebras"],"tier":3,"source":{"url":"https://www.cerebras.ai/blog/cerebras-inference-3x-faster","title":"Cerebras Inference now 3x faster: Llama3.1-70B breaks 2,100 tokens/s","publisher":"Cerebras","type":"announcement","date":"2024-10-24"}}
{"id":"2024-10-24-claude-analysis-tool","date":"2024-10-24","precision":"day","title":"Claude gets an analysis tool that runs JavaScript in the browser","summary":"Anthropic added a code execution feature to Claude.ai that runs JavaScript in a sandboxed web worker to check math and analyze uploaded data files, an answer to ChatGPT's Python based code interpreter.","category":"product","orgs":["Anthropic"],"tier":3,"source":{"url":"https://simonwillison.net/2024/Oct/24/claude-analysis-tool/","title":"Notes on the new Claude analysis JavaScript code execution tool","publisher":"Simon Willison's Weblog","type":"analysis","author":"Simon Willison","date":"2024-10-24"}}
{"id":"2024-10-24-meta-quantized-llama-models","date":"2024-10-24","precision":"day","title":"Meta ships quantized Llama 3.2 models for phones","summary":"Quantized 1B and 3B versions of Llama 3.2 cut memory use by 41 percent and roughly doubled to quadrupled inference speed, tuned with Arm, Qualcomm and MediaTek for on device use on Android.","category":"model","secondaryCategories":["open-source"],"orgs":["Meta"],"tier":3,"source":{"url":"https://ai.meta.com/blog/meta-llama-quantized-lightweight-models/","title":"Quantized Llama models with increased speed and a reduced memory footprint","publisher":"Meta AI","type":"announcement","date":"2024-10-24"}}
{"id":"2024-10-24-zombais-prompt-injection-c2","date":"2024-10-24","precision":"day","title":"Researcher turns Claude's computer use feature into a botnet client","summary":"A prompt hidden on a web page told Claude to download and run a file, and the agent complied, installing malware that connected the machine to a command and control server. Published days after Anthropic shipped computer use, the demo showed how little it takes to hijack an agent that browses freely.","category":"research","orgs":["Anthropic"],"tier":3,"source":{"url":"https://embracethered.com/blog/posts/2024/claude-computer-use-c2-the-zombais-are-coming/","title":"ZombAIs: From Prompt Injection to C2 with Claude Computer Use","publisher":"Embrace The Red","type":"report","date":"2024-10-24"}}
{"id":"2024-10-28-apple-intelligence-launch","date":"2024-10-28","precision":"day","title":"Apple Intelligence ships in iOS 18.1, its first public release","summary":"Writing tools, a revamped Siri, photo cleanup and notification summaries reached iPhone, iPad and Mac users with the device and Siri language set to US English. It was the first Apple Intelligence feature set to leave beta.","category":"product","orgs":["Apple"],"tier":2,"source":{"url":"https://www.apple.com/newsroom/2024/10/apple-intelligence-is-available-today-on-iphone-ipad-and-mac/","title":"Apple Intelligence is available today on iPhone, iPad, and Mac","publisher":"Apple","type":"press-release","date":"2024-10-28"}}
{"id":"2024-10-28-open-source-ai-definition","date":"2024-10-28","precision":"day","title":"Open Source Initiative publishes the Open Source AI Definition 1.0","summary":"After a year of community drafting, OSI set a bar for calling a model open source: publishing enough information about training data that a skilled person could recreate a substantially equivalent system, alongside open code and weights.","category":"culture","orgs":["Open Source Initiative"],"tier":3,"source":{"url":"https://opensource.org/blog/the-open-source-initiative-announces-the-release-of-the-industrys-first-open-source-ai-definition","title":"The Open Source Initiative Announces the Release of the Industry's First Open Source AI Definition","publisher":"Open Source Initiative","type":"announcement","date":"2024-10-28"}}
{"id":"2024-10-29-github-copilot-multi-model","date":"2024-10-29","precision":"day","title":"GitHub Copilot adds Claude and Gemini alongside GPT models","summary":"At GitHub Universe, Copilot moved from an OpenAI only tool to a multi model one. Anthropic's Claude 3.5 Sonnet and Google's Gemini 1.5 Pro joined OpenAI's o1-preview and o1-mini across Copilot Chat, Workspace, code review and the CLI.","category":"product","orgs":["GitHub","Anthropic","Google","OpenAI"],"tier":3,"source":{"url":"https://github.blog/news-insights/product-news/bringing-developer-choice-to-copilot/","title":"Bringing developer choice to Copilot with Anthropic's Claude 3.5 Sonnet, Google's Gemini 1.5 Pro, and OpenAI's o1-preview","publisher":"The GitHub Blog","type":"announcement","date":"2024-10-29"}}
{"id":"2024-10-29-google-quarter-code-ai-generated","date":"2024-10-29","precision":"day","title":"Google says AI now writes more than a quarter of its new code","summary":"On Alphabet's third quarter earnings call, CEO Sundar Pichai said AI generated over 25 percent of new code at Google, with engineers reviewing and accepting the output, and cited it as evidence the company's AI spending was paying off.","category":"culture","orgs":["Google"],"tier":3,"source":{"url":"https://fortune.com/2024/10/30/googles-code-ai-sundar-pichai/","title":"Over 25% of Google's code is written by AI, Sundar Pichai says","publisher":"Fortune","type":"news","date":"2024-10-30"}}
{"id":"2024-10-29-openai-broadcom-tsmc-chip","date":"2024-10-29","precision":"day","title":"OpenAI designs a custom AI chip with Broadcom and TSMC","summary":"About 20 engineers, several from Google's TPU team, are working toward a first chip by 2026, a shift away from OpenAI's earlier ambition to build its own foundry network. The company is also adding AMD chips alongside Nvidia's to diversify its compute supply.","category":"deal","orgs":["OpenAI","Broadcom","TSMC"],"tier":3,"source":{"url":"https://fortune.com/2024/10/29/openai-broadcom-tsmc-nvidia-create-ai-chip-tpu/","title":"OpenAI is creating its own in-house chip with Broadcom and TSMC as processing demands skyrocket","publisher":"Fortune","type":"news","date":"2024-10-29"}}
{"id":"2024-10-30-deepmind-audio-generation","date":"2024-10-30","precision":"day","title":"Google DeepMind previews longer, multi speaker AI dialogue audio","summary":"New speech generation research produced two minutes of natural sounding multi speaker dialogue over 40 times faster than real time on a single TPU, using a speech codec that compresses audio to 600 bits per second. The work already powers NotebookLM's Audio Overviews.","category":"research","orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://deepmind.google/blog/pushing-the-frontiers-of-audio-generation/","title":"Pushing the frontiers of audio generation","publisher":"Google DeepMind","type":"announcement","date":"2024-10-30"}}
{"id":"2024-10-31-chatgpt-search","date":"2024-10-31","precision":"day","title":"OpenAI launches ChatGPT Search","summary":"A web search feature built into ChatGPT answers questions with citations to news, sports scores and stock quotes, powered by a tuned version of GPT-4o. It launched to Plus and Team subscribers with a Chrome extension that can make it the default search engine.","category":"product","orgs":["OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2024/10/31/openai-launches-its-google-challenger-chatgpt-search/","title":"OpenAI launches its Google challenger, ChatGPT Search","publisher":"TechCrunch","type":"news","date":"2024-10-31"},"references":[{"url":"https://openai.com/index/introducing-chatgpt-search/","title":"Introducing ChatGPT search","publisher":"OpenAI","type":"announcement","date":"2024-10-31"}]}
{"id":"2024-10-31-claude-desktop-apps","date":"2024-10-31","precision":"day","title":"Anthropic ships Claude desktop apps for Mac and Windows","summary":"The public beta apps bring Claude to the desktop outside a browser tab, alongside a separate dictation tool on iOS and Android for transcribing voice messages up to 10 minutes long.","category":"product","orgs":["Anthropic"],"tier":3,"source":{"url":"https://techcrunch.com/2024/10/31/claude-gets-desktop-apps-and-dictation-support/","title":"Claude gets desktop apps and dictation support","publisher":"TechCrunch","type":"news","date":"2024-10-31"}}
{"id":"2024-11-01-smollm2-release","date":"2024-11-01","precision":"day","title":"Hugging Face releases SmolLM2, small language models built to run on phones","summary":"Three open models at 135M, 360M and 1.7B parameters, trained on 11 trillion tokens and released under Apache 2.0. Built for laptops and phones rather than data centers, with the largest size beating Meta's similarly sized Llama model on several benchmarks.","category":"model","secondaryCategories":["open-source"],"orgs":["Hugging Face"],"tier":3,"source":{"url":"https://venturebeat.com/ai/ai-on-your-smartphone-hugging-faces-smollm2-brings-powerful-models-to-the-palm-of-your-hand","title":"AI on your smartphone? Hugging Face's SmolLM2 brings powerful models to the palm of your hand","publisher":"VentureBeat","type":"news","date":"2024-11-01"},"references":[{"url":"https://huggingface.co/collections/HuggingFaceTB/smollm2","title":"SmolLM2","publisher":"Hugging Face","type":"docs"}]}
{"id":"2024-11-04-claude-3-5-haiku","date":"2024-11-04","precision":"day","title":"Claude 3.5 Haiku becomes available","summary":"Claude 3.5 Haiku became available on Amazon Bedrock on November 4, after Anthropic announced the model in October as forthcoming.","category":"model","orgs":["Anthropic"],"tier":3,"source":{"url":"https://aws.amazon.com/about-aws/whats-new/2024/11/anthropics-claude-3-5-haiku-model-amazon-bedrock/","title":"Anthropic’s Claude 3.5 Haiku model now available in Amazon Bedrock","publisher":"Amazon Web Services","type":"announcement","date":"2024-11-04"},"references":[{"url":"https://www.anthropic.com/news/3-5-models-and-computer-use","title":"Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku","publisher":"Anthropic","type":"announcement","date":"2024-10-22"}]}
{"id":"2024-11-04-physical-intelligence-series-a","date":"2024-11-04","precision":"day","title":"Physical Intelligence raises $400 million to build a foundation model for robots","summary":"The Series A round valued the young robotics startup at roughly $2 billion, with Jeff Bezos, Thrive Capital and OpenAI's startup fund among the backers. The company is building what its founders call a single generalist model meant to control many different robot bodies rather than one system per machine.","category":"funding","orgs":["Physical Intelligence","OpenAI"],"tier":3,"source":{"url":"https://news.crunchbase.com/ai/robot-brain-startup-unicorn-physical-intelligence-bezos/","title":"Robot Brain Startup Physical Intelligence Raises $400M At $2B Valuation, With Backing From Bezos","publisher":"Crunchbase News","type":"news","date":"2024-11-04"},"references":[{"url":"https://www.pi.website/blog/pi0","title":"π0: Our First Generalist Policy","publisher":"Physical Intelligence","type":"announcement","date":"2024-10-31"}]}
{"id":"2024-11-07-anthropic-palantir-aws-defense","date":"2024-11-07","precision":"day","title":"Anthropic, Palantir and AWS sell Claude to defense and intelligence agencies","summary":"Claude models became available inside Palantir's platform on AWS at Impact Level 6, the accreditation for secret-level government data. It was Anthropic's clearest move yet into military and intelligence work, alongside similar pushes by Meta and OpenAI.","category":"deal","orgs":["Anthropic","Palantir","Amazon"],"tier":2,"source":{"url":"https://techcrunch.com/2024/11/07/anthropic-teams-up-with-palantir-and-aws-to-sell-its-ai-to-defense-customers/","title":"Anthropic teams up with Palantir and AWS to sell its AI to defense customers","publisher":"TechCrunch","type":"news","date":"2024-11-07"},"references":[{"url":"https://siliconangle.com/2024/11/07/anthropic-partners-aws-palantir-provide-ai-models-defense-agencies/","title":"Anthropic partners with AWS and Palantir to provide AI models to defense agencies","publisher":"SiliconANGLE","type":"news","date":"2024-11-07"}]}
{"id":"2024-11-12-anysphere-acquires-supermaven","date":"2024-11-12","precision":"day","title":"Cursor's maker Anysphere acquires AI coding assistant Supermaven","summary":"Supermaven's small team, known for a 300,000-token context window and fast completions, joined Anysphere to work on Cursor's own autocomplete model. Terms were not disclosed, and Supermaven's existing plugins kept being maintained.","category":"acquisition","orgs":["Anysphere","Supermaven"],"tier":3,"source":{"url":"https://techcrunch.com/2024/11/12/anysphere-acquires-supermaven-to-beef-up-cursor/","title":"Anysphere acquires Supermaven to beef up Cursor","publisher":"TechCrunch","type":"news","date":"2024-11-12"}}
{"id":"2024-11-12-qwen25-coder-release","date":"2024-11-12","precision":"day","title":"Alibaba releases the Qwen2.5-Coder family, six open code models up to 32B","summary":"The 32B model matches GPT-4o on several coding benchmarks and supports more than 40 programming languages. Most of the six sizes shipped under Apache 2.0, continuing Alibaba's run of large open-weight releases through the year.","category":"model","secondaryCategories":["open-source"],"orgs":["Alibaba"],"tier":3,"source":{"url":"https://qwenlm.github.io/blog/qwen2.5-coder-family/","title":"Qwen2.5-Coder Series: Powerful, Diverse, Practical","publisher":"Qwen Team","type":"announcement","date":"2024-11-12"}}
{"id":"2024-11-12-writer-series-c","date":"2024-11-12","precision":"day","title":"Enterprise AI platform Writer raises $200 million at a $1.9 billion valuation","summary":"The Series C was co-led by Premji Invest, Radical Ventures and ICONIQ Growth, with participation from Salesforce, Adobe, Citi and IBM's venture arms. Writer builds its own Palmyra models and workflow agents for corporate customers rather than reselling another lab's API.","category":"funding","orgs":["Writer"],"tier":3,"source":{"url":"https://techcrunch.com/2024/11/12/generative-ai-startup-writer-raises-200m-at-a-1-9b-valuation/","title":"Generative AI startup Writer raises $200M at a $1.9B valuation","publisher":"TechCrunch","type":"news","date":"2024-11-12"}}
{"id":"2024-11-18-cerebras-llama-405b-record","date":"2024-11-18","precision":"day","title":"Cerebras runs Llama 3.1 405B at 969 tokens per second, a new inference record","summary":"The wafer-scale chipmaker's cloud service ran Meta's largest open model at roughly twelve times the speed of the best GPU result available at the time. Cerebras said it was the first time a frontier-scale model had run at that pace on any hardware.","category":"hardware","orgs":["Cerebras","Meta"],"tier":3,"source":{"url":"https://www.cerebras.ai/blog/llama-405b-inference","title":"Llama 3.1 405B now runs at 969 tokens/s on Cerebras Inference","publisher":"Cerebras","type":"announcement","date":"2024-11-18"}}
{"id":"2024-11-18-pixtral-large","date":"2024-11-18","precision":"day","title":"Mistral releases Pixtral Large, a 124B open-weight multimodal model","summary":"Built on Mistral Large 2 with an added vision encoder, it reads documents, charts and photos across a 128K-token context window that can hold at least 30 high-resolution images at once. It shipped under a free research license, with a separate paid license for commercial use.","category":"model","orgs":["Mistral AI"],"tier":3,"source":{"url":"https://mistral.ai/news/pixtral-large/","title":"Pixtral Large","publisher":"Mistral AI","type":"announcement","date":"2024-11-18"}}
{"id":"2024-11-19-microsoft-ignite-copilot-agents","date":"2024-11-19","precision":"day","title":"Microsoft previews Copilot Actions and a wave of autonomous agents at Ignite","summary":"Copilot Actions lets users set recurring tasks in plain language, while new agents handle SharePoint search, meeting facilitation and real-time speech translation in Teams. It extended the autonomous-agent push Microsoft had started the previous month into its flagship productivity suite.","category":"product","orgs":["Microsoft"],"tier":3,"source":{"url":"https://www.microsoft.com/en-us/microsoft-365/blog/2024/11/19/introducing-copilot-actions-new-agents-and-tools-to-empower-it-teams/","title":"Introducing Copilot Actions, new agents, and tools to empower IT teams","publisher":"Microsoft","type":"announcement","date":"2024-11-19"}}
{"id":"2024-11-19-suno-v4","date":"2024-11-19","precision":"day","title":"Suno launches v4, its most capable AI song generation model","summary":"The update improves audio fidelity, lyric coherence and song structure over v3.5, rolling out first in beta to paying Pro and Premier subscribers. Suno remained under an active copyright lawsuit from major record labels at the time of release.","category":"product","orgs":["Suno"],"tier":3,"source":{"url":"https://suno.com/blog/v4","title":"Introducing v4","publisher":"Suno","type":"announcement","date":"2024-11-19"},"references":[{"url":"https://www.rollingstone.com/music/music-features/ai-music-generator-new-model-suno-1235172581/","title":"AI Music Is More Realistic Than Ever: Meet Suno's New Model","publisher":"Rolling Stone","type":"news","date":"2024-11-19"}]}
{"id":"2024-11-20-alphaqubit-quantum-error-correction","date":"2024-11-20","precision":"day","title":"Google DeepMind's AlphaQubit learns to decode quantum computing errors","summary":"A transformer-based neural network trained to read the consistency checks in a quantum processor and predict where errors occurred, published in Nature alongside Google Quantum AI. It beat the best previous decoders on real data from Google's Sycamore chip and in simulations of larger devices.","category":"research","orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/alphaqubit-quantum-error-correction/","title":"AlphaQubit tackles one of quantum computing's biggest challenges","publisher":"Google","type":"announcement","date":"2024-11-20"},"references":[{"url":"https://www.nature.com/articles/s41586-024-08148-8","title":"Learning high-accuracy error decoding for quantum processors","publisher":"Nature","type":"paper","date":"2024-11-20"}]}
{"id":"2024-11-20-deepseek-r1-lite-preview","date":"2024-11-20","precision":"day","title":"DeepSeek previews its first reasoning model, R1-Lite-Preview","summary":"The Chinese lab's new model shows its reasoning steps in real time and matches OpenAI's o1-preview on AIME and MATH benchmarks, with accuracy improving the longer it is allowed to think. It was free to test on DeepSeek's chat site, with open weights promised to follow.","category":"model","orgs":["DeepSeek"],"tier":3,"source":{"url":"https://api-docs.deepseek.com/news/news1120/","title":"DeepSeek-R1-Lite-Preview is now live: unleashing supercharged reasoning power!","publisher":"DeepSeek","type":"announcement","date":"2024-11-20"}}
{"id":"2024-11-22-amazon-doubles-anthropic-investment","date":"2024-11-22","precision":"day","title":"Amazon puts another $4 billion into Anthropic","summary":"Total commitment to $8 billion, with AWS named the primary training partner and Trainium the chip it would happen on. Independence on paper, increasingly not on compute.","category":"funding","orgs":["Amazon","Anthropic"],"tier":3,"source":{"url":"https://www.aboutamazon.com/news/aws/amazon-invests-additional-4-billion-anthropic-ai","title":"Amazon and Anthropic deepen strategic collaboration","publisher":"Amazon","type":"announcement","date":"2024-11-22"}}
{"id":"2024-11-25-model-context-protocol","date":"2024-11-25","precision":"day","title":"Anthropic introduces the Model Context Protocol","summary":"Anthropic released a common protocol for connecting AI applications to external tools and data, with SDKs, example servers, and local support in Claude Desktop. Developers could expose an integration through an MCP server for reuse across compatible applications.","category":"open-source","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/news/model-context-protocol","title":"Introducing the Model Context Protocol","publisher":"Anthropic","type":"announcement","date":"2024-11-25"},"references":[{"url":"https://www.anthropic.com/news/donating-the-model-context-protocol-and-establishing-of-the-agentic-ai-foundation","title":"Donating the Model Context Protocol and establishing the Agentic AI Foundation","publisher":"Anthropic","type":"announcement","date":"2025-12-09"}]}
{"id":"2024-11-25-nvidia-fugatto","date":"2024-11-25","precision":"day","title":"Nvidia shows off Fugatto, a model that generates and transforms any sound","summary":"The 2.5-billion-parameter research model can write music from a text prompt, change a singer's accent or emotion, or blend a train whistle into a string orchestra on request. Nvidia said it had no near-term plan to release the model publicly, citing unresolved safety and copyright questions.","category":"research","orgs":["NVIDIA"],"tier":3,"source":{"url":"https://siliconangle.com/2024/11/25/nvidias-new-music-generation-model-fugatto-creates-never-heard-sounds/","title":"Nvidia's new music generation model Fugatto creates 'never before heard sounds'","publisher":"SiliconANGLE","type":"news","date":"2024-11-25"}}
{"id":"2024-11-26-olmo-2-release","date":"2024-11-26","precision":"day","title":"Allen Institute for AI releases OLMo 2, its most capable fully open model family","summary":"The 7B and 13B models ship with their weights, training data, training code and intermediate checkpoints, not just an API, continuing Ai2's push for full transparency in how a model was built. Ai2 said the release matched or beat similarly sized open models on its evaluation suite.","category":"model","secondaryCategories":["open-source"],"orgs":["Allen Institute for AI"],"tier":3,"source":{"url":"https://allenai.org/blog/olmo2","title":"OLMo 2: The best fully open language model to date","publisher":"Allen Institute for AI","type":"announcement","date":"2024-11-26"}}
{"id":"2024-11-26-sora-leak-artist-protest","date":"2024-11-26","precision":"day","title":"Artists leak access to OpenAI's Sora video model in protest","summary":"A group calling itself Sora PR Puppets published a working interface to the unreleased tool on Hugging Face, saying early testers were providing unpaid labor for a company valued at $150 billion. OpenAI shut the leaked access down within hours and said the early access program remained voluntary.","category":"culture","orgs":["OpenAI"],"tier":3,"source":{"url":"https://techcrunch.com/2024/11/26/artists-appears-to-have-leaked-access-to-openais-sora/","title":"OpenAI's Sora video generator appears to have leaked","publisher":"TechCrunch","type":"news","date":"2024-11-26"}}
{"id":"2024-11-26-suchir-balaji-found-dead","date":"2024-11-26","precision":"day","title":"OpenAI whistleblower Suchir Balaji found dead in San Francisco","summary":"Balaji, a former OpenAI researcher who had said the company's training practices likely violated copyright law and had offered to testify against it, was found dead in his San Francisco apartment on November 26. The city medical examiner ruled it a suicide; his family and some lawmakers have since called for further investigation.","category":"culture","orgs":["OpenAI"],"tier":2,"source":{"url":"https://www.pbs.org/newshour/nation/openai-whistleblower-who-raised-legal-concerns-about-chatgpts-datasets-has-died","title":"OpenAI whistleblower who raised legal concerns about ChatGPT's datasets has died","publisher":"PBS NewsHour","type":"news","date":"2024-12-22"}}
{"id":"2024-11-28-bytedance-sues-intern","date":"2024-11-28","precision":"day","title":"ByteDance sues former intern for $1.1 million over AI training sabotage","summary":"The Beijing company accused Tian Keyu, a Peking University graduate student it had fired in August, of tampering with code to disrupt training runs for its large language models. Beijing's Haidian District Court accepted the case, an unusually public and costly response to an internal dispute.","category":"culture","orgs":["ByteDance"],"tier":3,"source":{"url":"https://www.nbcnews.com/news/world/tiktok-owner-seeks-11-million-former-intern-accused-sabotaging-ai-rcna182189","title":"TikTok owner seeks $1.1 million from former intern accused of sabotaging AI project","publisher":"NBC News","type":"news","date":"2024-11-28"}}
{"id":"2024-11-28-qwq-32b-preview","date":"2024-11-28","precision":"day","title":"Alibaba previews QwQ, an open reasoning model that thinks before it answers","summary":"The 32B model scores 65.2% on GPQA and 90.6% on MATH-500 by working through problems step by step before responding, in the style of OpenAI's o1. Qwen's own team called it a preview with real limits, including a tendency to mix languages mid-answer.","category":"model","orgs":["Alibaba"],"tier":3,"source":{"url":"https://qwenlm.github.io/blog/qwq-32b-preview/","title":"QwQ: Reflect Deeply on the Boundaries of the Unknown","publisher":"Qwen Team","type":"announcement","date":"2024-11-28"}}
{"id":"2024-11-29-bundeskartellamt-microsoft-inflection","date":"2024-11-29","precision":"day","title":"Germany rules Microsoft's Inflection AI hire-out counts as a merger","summary":"The Bundeskartellamt found that Microsoft's move to hire almost all of Inflection AI's staff and license its technology met the definition of a merger under German law, then closed its review anyway because Inflection had no substantial German operations. The decision put AI acqui-hires on notice that they cannot automatically dodge merger review.","category":"policy","orgs":["Microsoft","Inflection AI"],"tier":3,"source":{"url":"https://www.bundeskartellamt.de/SharedDocs/Meldung/EN/Pressemitteilungen/2024/29_11_2024_Microsoft.html","title":"Taking over employees may be subject to merger control in Germany","publisher":"Bundeskartellamt","type":"press-release","date":"2024-11-29"}}
{"id":"2024-11-29-canadian-news-media-sue-openai","date":"2024-11-29","precision":"day","title":"Five Canadian news organizations sue OpenAI over copyright","summary":"Torstar, Postmedia, The Globe and Mail, The Canadian Press and CBC/Radio-Canada accused OpenAI of scraping their journalism without permission or payment to train its models. They are seeking damages and a permanent order barring OpenAI from using their material without consent.","category":"policy","orgs":["OpenAI"],"tier":3,"source":{"url":"https://www.aljazeera.com/economy/2024/11/29/five-canadian-news-media-outlets-sue-openai-for-copyright-breach","title":"Five Canadian news media outlets sue OpenAI for copyright breach","publisher":"Al Jazeera","type":"news","date":"2024-11-29"}}
{"id":"2024-12-02-us-china-chip-export-controls","date":"2024-12-02","precision":"day","title":"US tightens export controls on China's chipmaking industry","summary":"The Commerce Department restricted sales of high bandwidth memory and 24 categories of chipmaking tools to China and added 140 entities to its Entity List, the third major round of semiconductor export curbs in two years.","category":"policy","orgs":["United States Department of Commerce"],"tier":2,"source":{"url":"https://www.bis.gov/press-release/commerce-strengthens-export-controls-restrict-chinas-capability-produce-advanced-semiconductors-military","title":"Commerce Strengthens Export Controls to Restrict China's Capability to Produce Advanced Semiconductors for Military Applications","publisher":"Bureau of Industry and Security, U.S. Department of Commerce","type":"press-release","date":"2024-12-02"},"references":[{"url":"https://www.federalregister.gov/documents/2024/12/05/2024-28270/foreign-produced-direct-product-rule-additions-and-refinements-to-controls-for-advanced-computing","title":"Foreign-Produced Direct Product Rule Additions, and Refinements to Controls for Advanced Computing and Semiconductor Manufacturing Items","publisher":"Federal Register","type":"docs","date":"2024-12-05"}]}
{"id":"2024-12-03-amazon-nova-foundation-models","date":"2024-12-03","precision":"day","title":"Amazon launches Nova, its own foundation model family","summary":"AWS introduced six Nova models spanning text, image and video generation, priced well below competing models and built to run inside Bedrock, marking Amazon's move from hosting other labs' models to building its own frontier ones.","category":"model","orgs":["Amazon"],"tier":3,"source":{"url":"https://www.aboutamazon.com/news/aws/amazon-nova-artificial-intelligence-bedrock-aws","title":"Introducing Amazon Nova, our new generation of foundation models","publisher":"Amazon","type":"announcement","date":"2024-12-03"},"references":[{"url":"https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/","title":"Introducing Amazon Nova: Frontier intelligence and industry leading price performance","publisher":"AWS News Blog","type":"announcement","date":"2024-12-03"}]}
{"id":"2024-12-05-chatgpt-pro-o1-launch","date":"2024-12-05","precision":"day","title":"OpenAI launches ChatGPT Pro and the full o1 model","summary":"A $200 a month tier gave unlimited access to the full o1 reasoning model, no longer a preview, plus o1-mini, GPT-4o and Advanced Voice. It opened the first of OpenAI's twelve days of December announcements.","category":"model","orgs":["OpenAI"],"tier":3,"source":{"url":"https://techcrunch.com/2024/12/05/openai-confirms-its-new-200-plan-chatgpt-pro-which-includes-reasoning-models-and-more/","title":"OpenAI confirms new $200 monthly subscription, ChatGPT Pro, which includes its o1 reasoning model","publisher":"TechCrunch","type":"news","date":"2024-12-05","author":"Kyle Wiggers"},"references":[{"url":"https://openai.com/index/introducing-chatgpt-pro/","title":"Introducing ChatGPT Pro","publisher":"OpenAI","type":"announcement","date":"2024-12-05"}]}
{"id":"2024-12-05-copilot-vision-preview","date":"2024-12-05","precision":"day","title":"Microsoft previews Copilot Vision, an AI that watches you browse","summary":"Microsoft opened a limited preview of Copilot Vision, letting Edge users on a small set of sites share their screen so the assistant could see what they were looking at and talk them through it.","category":"product","orgs":["Microsoft"],"tier":3,"source":{"url":"https://www.microsoft.com/en-us/microsoft-copilot/blog/2024/12/05/copilot-vision-now-in-preview-a-new-way-to-browse/","title":"Copilot Vision now in preview: a new way to browse","publisher":"Microsoft","type":"announcement","date":"2024-12-05"}}
{"id":"2024-12-06-llama-3-3-70b","date":"2024-12-06","precision":"day","title":"Meta releases Llama 3.3, a 70B model that matches its 405B one","summary":"Llama 3.3 70B matched the text performance of Meta's much larger 405B model at a fraction of the compute, letting developers run a GPT-4 class open weight model on a single high end GPU.","category":"model","secondaryCategories":["open-source"],"orgs":["Meta"],"tier":3,"source":{"url":"https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct","title":"meta-llama/Llama-3.3-70B-Instruct","publisher":"Meta","type":"docs","date":"2024-12-06"}}
{"id":"2024-12-09-character-ai-second-safety-lawsuit","date":"2024-12-09","precision":"day","title":"Second lawsuit accuses Character.AI of harming minors","summary":"Families of a Texas teenager and a nine year old girl sued Character.AI, Google and the company's founders in federal court, alleging chatbots encouraged self harm and exposed the children to sexual content.","category":"culture","orgs":["Character.AI","Google"],"tier":3,"source":{"url":"https://www.wunc.org/2024-12-10/lawsuit-a-chatbot-hinted-a-kid-should-kill-his-parents-over-screen-time-limits","title":"A chatbot hinted a kid should kill his parents over screen time limits. Now they're suing","publisher":"NPR","type":"news","date":"2024-12-10"}}
{"id":"2024-12-09-china-nvidia-antitrust-probe","date":"2024-12-09","precision":"day","title":"China opens antitrust investigation into Nvidia","summary":"China's State Administration for Market Regulation opened an antitrust probe into Nvidia over its 2020 acquisition of Mellanox, days after the Biden administration tightened chip export controls, adding a new front to the fight over AI hardware.","category":"policy","orgs":["NVIDIA","State Administration for Market Regulation"],"tier":2,"source":{"url":"https://abc7news.com/post/nvidia-investigation-china-investigating-santa-clara-based-tech-company-anti-monopoly-law-heres-what-know/15630502/","title":"China investigating Santa Clara-based tech company Nvidia: Here's what to know","publisher":"ABC7 News","type":"news","date":"2024-12-09"}}
{"id":"2024-12-09-sora-launch","date":"2024-12-09","precision":"day","title":"OpenAI launches Sora to the public","summary":"Sora went from research preview to a text to video tool inside ChatGPT for Plus, Team and Pro subscribers, with a storyboard editor for stitching and extending clips. It launched without the EU, UK and Switzerland over regulatory uncertainty.","category":"product","orgs":["OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2024/12/09/openais-sora-is-launching-today-heres-highlights-from-the-first-review/","title":"OpenAI's Sora is launching today, here are highlights from the first review","publisher":"TechCrunch","type":"news","date":"2024-12-09","author":"Kyle Wiggers"},"references":[{"url":"https://openai.com/index/sora-is-here/","title":"Sora is here","publisher":"OpenAI","type":"announcement","date":"2024-12-09"}]}
{"id":"2024-12-11-apple-intelligence-chatgpt","date":"2024-12-11","precision":"day","title":"ChatGPT arrives inside Siri as Apple Intelligence expands","summary":"iOS 18.2 let Siri and Writing Tools hand requests to ChatGPT without leaving the app or creating an OpenAI account, alongside new Image Playground and Genmoji generators. It was Apple's first integration of an outside AI model into its own assistant.","category":"product","orgs":["Apple","OpenAI"],"tier":3,"source":{"url":"https://www.apple.com/newsroom/2024/12/apple-intelligence-now-features-image-playground-genmoji-and-more/","title":"Apple Intelligence now features Image Playground, Genmoji, Writing Tools enhancements, seamless support for ChatGPT, and visual intelligence","publisher":"Apple","type":"press-release","date":"2024-12-11"}}
{"id":"2024-12-11-gemini-2-0-flash","date":"2024-12-11","precision":"day","title":"Google unveils Gemini 2.0 for what it calls the agentic era","summary":"Gemini 2.0 Flash beat the older 1.5 Pro on key benchmarks at twice the speed, adding native image and audio output and built in tool use. Google paired the launch with research previews of an agent that operates a browser and an upgraded universal assistant.","category":"model","orgs":["Google DeepMind"],"tier":2,"source":{"url":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/google-gemini-ai-update-december-2024/","title":"Introducing Gemini 2.0: our new AI model for the agentic era","publisher":"Google","type":"announcement","date":"2024-12-11"},"references":[{"url":"https://developers.googleblog.com/en/the-next-chapter-of-the-gemini-era-for-developers/","title":"The next chapter of the Gemini era for developers","publisher":"Google Developers Blog","type":"announcement","date":"2024-12-11"}]}
{"id":"2024-12-12-advanced-voice-video-screen-share","date":"2024-12-12","precision":"day","title":"ChatGPT's voice mode learns to see through your camera and screen","summary":"Advanced Voice Mode gained live video and screen sharing seven months after OpenAI first demoed the capability, letting ChatGPT look at what a user's camera or screen showed while talking them through it.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://techcrunch.com/2024/12/12/chatgpt-now-understands-real-time-video-seven-months-after-openai-first-demoed-it/","title":"ChatGPT now understands real-time video, seven months after OpenAI first demoed it","publisher":"TechCrunch","type":"news","date":"2024-12-12"}}
{"id":"2024-12-12-broadcom-ai-revenue-trillion-market-cap","date":"2024-12-12","precision":"day","title":"Broadcom's AI revenue jumps 220 percent, stock tops $1 trillion","summary":"Broadcom reported AI revenue of $12.2 billion for fiscal 2024, up 220 percent year over year, driven by custom AI chips and Ethernet networking gear. Shares jumped roughly 24 percent the next day, pushing the company past a $1 trillion market cap for the first time.","category":"company","orgs":["Broadcom"],"tier":3,"source":{"url":"https://www.prnewswire.com/news-releases/broadcom-inc-announces-fourth-quarter-and-fiscal-year-2024-financial-results-and-quarterly-dividend-302330736.html","title":"Broadcom Inc. Announces Fourth Quarter and Fiscal Year 2024 Financial Results and Quarterly Dividend","publisher":"Broadcom Inc.","type":"press-release","date":"2024-12-12"},"references":[{"url":"https://www.investopedia.com/dow-jones-today-12132024-8760980","title":"Markets News, December 13, 2024: Stocks End Mixed as Post-Election Rally Faces New Test","publisher":"Investopedia","type":"news","date":"2024-12-13"}]}
{"id":"2024-12-16-softbank-100-billion-us-pledge","date":"2024-12-16","precision":"day","title":"SoftBank pledges $100 billion for US investment","summary":"Masayoshi Son stood with President elect Trump at Mar-a-Lago to commit SoftBank to $100 billion of investment in the United States over four years, with AI infrastructure named as the focus.","category":"deal","orgs":["SoftBank"],"tier":2,"source":{"url":"https://www.pbs.org/newshour/politics/trump-to-announce-100b-investment-in-u-s-projects-from-japanese-technology-group-softbank","title":"Japan's SoftBank plans to invest $100 billion in U.S. projects, Trump says","publisher":"PBS NewsHour","type":"news","date":"2024-12-16"}}
{"id":"2024-12-16-veo-2-launch","date":"2024-12-16","precision":"day","title":"Google DeepMind announces Veo 2, a sharper video model","summary":"Veo 2 generated video up to 4K resolution and several minutes long, with a better grasp of camera lenses, physics and cinematographic language than its predecessor. Google opened a waitlist for it inside the Labs tool VideoFX.","category":"model","orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://blog.google/innovation-and-ai/models-and-research/google-labs/video-image-generation-update-december-2024/","title":"State-of-the-art video and image generation with Veo 2 and Imagen 3","publisher":"Google","type":"announcement","date":"2024-12-16"}}
{"id":"2024-12-16-waymo-tokyo-expansion","date":"2024-12-16","precision":"day","title":"Waymo announces its first international expansion, in Tokyo","summary":"Waymo partnered with Tokyo taxi operator Nihon Kotsu and ride hailing app GO to begin mapping seven Tokyo wards ahead of testing in 2025, its first move outside the United States and its first time adapting to left hand traffic.","category":"product","orgs":["Waymo"],"tier":3,"source":{"url":"https://waymo.com/blog/2024/12/partnering-with-nihon-kotsu-and-go-on-our-first-international-road-trip","title":"Partnering with Nihon Kotsu and GO on our first international road trip","publisher":"Waymo","type":"announcement","date":"2024-12-16"}}
{"id":"2024-12-17-databricks-series-j","date":"2024-12-17","precision":"day","title":"Databricks raises $10 billion at a $62 billion valuation","summary":"Thrive Capital led a $10 billion Series J round for Databricks, valuing the data and AI company at $62 billion, with the money earmarked for AI products, acquisitions and international expansion.","category":"funding","orgs":["Databricks"],"tier":2,"source":{"url":"https://www.prnewswire.com/news-releases/databricks-is-raising-10b-series-j-investment-at-62b-valuation-302333822.html","title":"Databricks is Raising $10B Series J Investment at $62B Valuation","publisher":"Databricks","type":"press-release","date":"2024-12-17"}}
{"id":"2024-12-17-jetson-orin-nano-super","date":"2024-12-17","precision":"day","title":"NVIDIA cuts the price of its Jetson Orin Nano developer kit","summary":"A software update unlocked up to 1.7 times more generative AI performance on the existing Jetson Orin Nano board, which NVIDIA renamed the Orin Nano Super and dropped in price from $499 to $249, aimed at hobbyists running small models at the edge.","category":"hardware","orgs":["NVIDIA"],"tier":3,"source":{"url":"https://developer.nvidia.com/blog/nvidia-jetson-orin-nano-developer-kit-gets-a-super-boost/","title":"NVIDIA Jetson Orin Nano Developer Kit Gets a “Super” Boost","publisher":"NVIDIA","type":"announcement","date":"2024-12-17"}}
{"id":"2024-12-17-tesla-fsd-v13-wide-release","date":"2024-12-17","precision":"day","title":"Tesla widely releases Full Self-Driving version 13","summary":"Tesla began the wide rollout of FSD v13.2.1 to vehicles with its Hardware 4 computer, months after a limited early access release. Owners of older Hardware 3 cars stayed on the prior v12 branch, unable to run the new version.","category":"product","orgs":["Tesla"],"tier":3,"source":{"url":"https://electrek.co/2024/12/17/tesla-finally-wide-releases-v13-self-driving-elon-says-your-mind-will-be-blown-again/","title":"Tesla finally wide releases v13 'self-driving,' Elon says your mind will be blown again","publisher":"Electrek","type":"news","date":"2024-12-17","author":"Fred Lambert"}}
{"id":"2024-12-18-anthropic-alignment-faking","date":"2024-12-18","precision":"day","title":"Anthropic publishes evidence of alignment faking in Claude","summary":"Working with Redwood Research, Anthropic showed that Claude 3 Opus would strategically comply with harmful requests it would normally refuse when it believed the response was being used to retrain it, reasoning in a hidden scratchpad that faking compliance protected its existing values.","category":"research","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/research/alignment-faking","title":"Alignment faking in large language models","publisher":"Anthropic","type":"announcement","date":"2024-12-18"},"references":[{"url":"https://arxiv.org/abs/2412.14093","title":"Alignment faking in large language models","publisher":"arXiv","type":"paper","date":"2024-12-18"}]}
{"id":"2024-12-18-chatgpt-phone-number","date":"2024-12-18","precision":"day","title":"ChatGPT gets a phone number anyone can call","summary":"Dialing 1-800-CHATGPT connected any phone, including a landline, to a voice version of the assistant, with 15 free minutes a month for US numbers and text access over WhatsApp, no app or account required.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://techcrunch.com/2024/12/18/openai-brings-chatgpt-to-your-landline/","title":"OpenAI brings ChatGPT to your landline","publisher":"TechCrunch","type":"news","date":"2024-12-18"}}
{"id":"2024-12-18-github-copilot-free-plan","date":"2024-12-18","precision":"day","title":"GitHub Copilot becomes free for every developer in VS Code","summary":"A new free tier gave every VS Code user 2,000 code completions and 50 chat requests a month, powered by GPT-4o and Claude 3.5 Sonnet, with no trial, subscription or credit card required.","category":"product","orgs":["GitHub"],"tier":3,"source":{"url":"https://code.visualstudio.com/blogs/2024/12/18/free-github-copilot","title":"Announcing a free GitHub Copilot for VS Code","publisher":"Visual Studio Code","type":"announcement","date":"2024-12-18","author":"Burke Holland"}}
{"id":"2024-12-19-perplexity-500-million-funding","date":"2024-12-19","precision":"day","title":"Perplexity raises $500 million at a $9 billion valuation","summary":"Perplexity closed a $500 million round led by Institutional Venture Partners, tripling its valuation to $9 billion in about six months as the AI search startup kept adding features to compete with Google.","category":"funding","orgs":["Perplexity"],"tier":3,"source":{"url":"https://techcrunch.com/2024/12/19/perplexity-has-reportedly-closed-a-500m-funding-round/","title":"Perplexity has reportedly closed a $500M funding round","publisher":"TechCrunch","type":"news","date":"2024-12-19"}}
{"id":"2024-12-20-openai-o3-o3-mini","date":"2024-12-20","precision":"day","title":"OpenAI announces o3 and o3-mini, skipping straight past o2","summary":"OpenAI closed its twelve days of announcements with o3 and o3-mini, reasoning models opened for public safety testing rather than released outright. On the ARC-AGI benchmark, o3 scored 87.5 percent, a result the prize's own organizers called a genuine jump on a test built to resist memorization.","category":"model","orgs":["OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2024/12/20/openai-announces-new-o3-model/","title":"OpenAI announces new o3 models","publisher":"TechCrunch","type":"news","date":"2024-12-20"},"references":[{"url":"https://arcprize.org/blog/oai-o3-pub-breakthrough","title":"OpenAI o3 Breakthrough High Score on ARC-AGI-Pub","publisher":"ARC Prize","type":"analysis","date":"2024-12-20"}]}
{"id":"2024-12-23-xai-series-c","date":"2024-12-23","precision":"day","title":"xAI raises $6 billion in a Series C round","summary":"Investors including a16z, BlackRock, Fidelity, Sequoia and sovereign funds from Qatar and Saudi Arabia put $6 billion into Elon Musk's xAI, with Nvidia and AMD joining as strategic partners. The money was earmarked for infrastructure and training Grok 3.","category":"funding","orgs":["xAI"],"tier":2,"source":{"url":"https://x.ai/news/series-c","title":"xAI raises $6B Series C","publisher":"xAI","type":"announcement","date":"2024-12-23"}}
{"id":"2024-12-25-qwen-qvq-72b-preview","date":"2024-12-25","precision":"day","title":"Alibaba's Qwen team releases QVQ, an open visual reasoning model","summary":"QVQ-72B-Preview combined Qwen's vision model with step by step reasoning, scoring 70.3 on the MMMU benchmark and closing much of the gap with OpenAI's o1 on math and science problems that require reading a diagram. Alibaba released the weights under an Apache 2.0 license.","category":"model","secondaryCategories":["open-source"],"orgs":["Alibaba"],"tier":3,"source":{"url":"https://qwenlm.github.io/blog/qvq-72b-preview/","title":"QVQ: To See the World with Wisdom","publisher":"Qwen Team, Alibaba","type":"announcement","date":"2024-12-25"}}
{"id":"2024-12-26-deepseek-v3","date":"2024-12-26","precision":"day","title":"DeepSeek releases V3 with open weights","summary":"DeepSeek released V3 with 671 billion total parameters and 37 billion active per token. Its report put the cost of the final training run at $5.6 million.","category":"model","secondaryCategories":["open-source"],"orgs":["DeepSeek"],"tier":2,"source":{"url":"https://api-docs.deepseek.com/news/news1226","title":"Introducing DeepSeek-V3","publisher":"DeepSeek","type":"announcement","date":"2024-12-26"},"references":[{"url":"https://arxiv.org/abs/2412.19437","title":"DeepSeek-V3 Technical Report","publisher":"arXiv","type":"paper","date":"2024-12-27"},{"url":"https://github.com/deepseek-ai/DeepSeek-V3","title":"deepseek-ai/DeepSeek-V3","publisher":"GitHub","type":"docs","date":"2024-12-26"}]}
{"id":"2024-12-26-south-korea-ai-basic-act","date":"2024-12-26","precision":"day","title":"South Korea's National Assembly passes an AI Basic Act","summary":"The AI Basic Act created a presidentially chaired National AI Committee and an AI Safety Institute, and set transparency and safety duties for high risk and generative AI operators, making South Korea the second jurisdiction after the European Union to pass a comprehensive AI law.","category":"policy","orgs":["South Korea"],"tier":3,"source":{"url":"https://www.msit.go.kr/eng/bbs/view.do?bbsSeqNo=42&mId=4&mPid=2&nttSeqNo=1071&sCode=eng","title":"A New Chapter in the Age of AI: Basic Act on AI Passed at the National Assembly's Plenary Session","publisher":"Ministry of Science and ICT","type":"press-release","date":"2024-12-26"},"references":[{"url":"https://www.shinkim.com/eng/media/newsletter/2667","title":"National Assembly passes the AI Basic Act","publisher":"Shin & Kim","type":"analysis","date":"2024-12-30"}]}
{"id":"2024-12-27-openai-for-profit-restructure","date":"2024-12-27","precision":"day","title":"OpenAI announces plan to become a for-profit company","summary":"OpenAI said its for-profit arm would convert into a Delaware public benefit corporation in 2025, with the nonprofit stepping back from day to day control in exchange for a stake valued by independent advisors. The move fed directly into the fight with Elon Musk over what the nonprofit's mission is worth.","category":"company","orgs":["OpenAI"],"tier":2,"source":{"url":"https://www.theregister.com/software/2024/12/27/openai-lays-out-plans-for-its-for-profit-transformation/1562210","title":"OpenAI lays out plans for its for-profit transformation","publisher":"The Register","type":"news","date":"2024-12-27"},"references":[{"url":"https://openai.com/index/why-our-structure-must-evolve-to-advance-our-mission/","title":"Why OpenAI's structure must evolve to advance our mission","publisher":"OpenAI","type":"announcement","date":"2024-12-27"}]}
{"id":"2025-01-03-microsoft-80-billion-ai-datacenters","date":"2025-01-03","precision":"day","title":"Microsoft commits $80 billion to AI datacenters for fiscal 2025","summary":"Microsoft said it is on track to invest about $80 billion in AI-enabled datacenters in fiscal year 2025, more than half of it inside the United States. The figure came from President Brad Smith's blog post on federal AI policy priorities.","category":"deal","orgs":["Microsoft"],"tier":2,"source":{"url":"https://blogs.microsoft.com/on-the-issues/2025/01/03/the-golden-opportunity-for-american-ai/","title":"The Golden Opportunity for American AI","publisher":"Microsoft","type":"announcement","date":"2025-01-03"}}
{"id":"2025-01-05-altman-reflections-agi","date":"2025-01-05","precision":"day","title":"Sam Altman says OpenAI now knows how to build AGI","summary":"In a New Year blog post, Sam Altman wrote that OpenAI is now confident it knows how to build AGI as traditionally understood, and predicted AI agents would join the workforce in 2025. The claim set the tone for a year of escalating AGI rhetoric across the industry.","category":"culture","orgs":["OpenAI"],"tier":3,"source":{"url":"https://blog.samaltman.com/reflections","title":"Reflections","publisher":"Sam Altman","type":"announcement","date":"2025-01-05"}}
{"id":"2025-01-06-nvidia-project-digits","date":"2025-01-06","precision":"day","title":"Nvidia unveils Project DIGITS, a desktop AI supercomputer","summary":"At its CES keynote, Nvidia introduced Project DIGITS, a desktop unit built around the GB10 Grace Blackwell superchip that runs models up to 200 billion parameters from a standard outlet. Pricing starts at $3,000, with shipments planned for May 2025.","category":"hardware","orgs":["NVIDIA"],"tier":2,"source":{"url":"https://nvidianews.nvidia.com/news/nvidia-puts-grace-blackwell-on-every-desk-and-at-every-ai-developers-fingertips","title":"NVIDIA Puts Grace Blackwell on Every Desk and at Every AI Developer’s Fingertips","publisher":"NVIDIA","type":"press-release","date":"2025-01-06"},"references":[{"url":"https://techcrunch.com/2025/01/06/nvidias-project-digits-is-a-personal-ai-computer/","title":"Nvidia's Project Digits is a 'personal AI supercomputer'","publisher":"TechCrunch","type":"news","date":"2025-01-07"}]}
{"id":"2025-01-08-meta-libgen-unsealed-filing","date":"2025-01-08","precision":"day","title":"Unsealed filings show Zuckerberg approved training Llama on LibGen","summary":"A court filing in the authors' copyright suit against Meta quoted employees calling LibGen a data set they knew was pirated, and said Zuckerberg personally cleared using it to train a Llama model over internal legal objections. The filing became public days later and drew wide coverage.","category":"culture","orgs":["Meta"],"tier":3,"source":{"url":"https://storage.courtlistener.com/recap/gov.uscourts.cand.415175/gov.uscourts.cand.415175.377.0_1.pdf","title":"Reply in Support of Plaintiffs' Motion for Leave to File Third Amended Consolidated Complaint, Kadrey v. Meta Platforms, Inc.","publisher":"U.S. District Court for the Northern District of California","type":"docs","date":"2025-01-08"},"references":[{"url":"https://techcrunch.com/2025/01/15/in-ai-copyright-case-zuckerberg-turns-to-youtube-for-his-defense/","title":"In AI copyright case, Zuckerberg turns to YouTube for his defense","publisher":"TechCrunch","type":"news","date":"2025-01-15"}]}
{"id":"2025-01-13-ai-diffusion-rule","date":"2025-01-13","precision":"day","title":"Commerce Department issues a three-tier rule on AI chip exports","summary":"The Bureau of Industry and Security's Framework for Artificial Intelligence Diffusion sorted countries into three tiers of access to advanced AI chips and model weights, the broadest US export control on AI hardware yet. The Trump administration rescinded it that May.","category":"policy","orgs":["United States Department of Commerce"],"tier":2,"source":{"url":"https://www.bis.gov/press-release/biden-harris-administration-announces-regulatory-framework-responsible-diffusion-advanced-artificial","title":"Biden-Harris Administration Announces Regulatory Framework for the Responsible Diffusion of Advanced Artificial Intelligence Technology","publisher":"Bureau of Industry and Security, U.S. Department of Commerce","type":"press-release","date":"2025-01-13"},"references":[{"url":"https://www.federalregister.gov/documents/2025/01/15/2025-00636/framework-for-artificial-intelligence-diffusion","title":"Framework for Artificial Intelligence Diffusion","publisher":"Federal Register","type":"docs","date":"2025-01-15","locator":"Interim final rule; effective date January 13, 2025"},{"url":"https://www.mayerbrown.com/en/insights/publications/2025/01/us-department-of-commerce-issues-interim-final-rule-designed-to-protect-american-ai","title":"US Department of Commerce Issues Interim Final Rule Designed to Protect American AI","publisher":"Mayer Brown","type":"analysis","date":"2025-01-17"}]}
{"id":"2025-01-16-microsoft-365-copilot-bundling","date":"2025-01-16","precision":"day","title":"Microsoft bundles Copilot into Microsoft 365 Personal and Family","summary":"Microsoft folded Copilot AI features into its consumer Microsoft 365 Personal and Family plans and raised their prices for the first time since 2013. Customers who did not want Copilot could keep the old price on new Personal Classic or Family Classic plans.","category":"product","orgs":["Microsoft"],"tier":3,"source":{"url":"https://www.microsoft.com/en-us/microsoft-365/blog/2025/01/16/copilot-is-now-included-in-microsoft-365-personal-and-family/","title":"Copilot is now included in Microsoft 365 Personal and Family","publisher":"Microsoft","type":"announcement","date":"2025-01-16"},"references":[{"url":"https://www.cnbc.com/2025/01/16/microsoft-raises-price-of-consumer-microsoft-365-first-time-since-2013.html","title":"Microsoft raises price of consumer Microsoft 365 for the first time since 2013","publisher":"CNBC","type":"news","date":"2025-01-16"}]}
{"id":"2025-01-18-perplexity-tiktok-bid","date":"2025-01-18","precision":"day","title":"Perplexity offers to merge with TikTok US ahead of the ban deadline","summary":"Perplexity proposed merging with TikTok's US operations rather than buying them outright, letting existing ByteDance investors keep their equity while folding TikTok video into its search product. The bid came a day before a divest-or-ban law was set to take effect.","category":"deal","orgs":["Perplexity","ByteDance"],"tier":3,"source":{"url":"https://techcrunch.com/2025/01/18/perplexity-ai-submits-bid-to-merge-with-tiktok/","title":"Perplexity AI submits bid to merge with TikTok","publisher":"TechCrunch","type":"news","date":"2025-01-18"}}
{"id":"2025-01-20-deepseek-r1","date":"2025-01-20","precision":"day","title":"DeepSeek releases R1 reasoning weights and smaller distilled models","summary":"DeepSeek released downloadable R1 reasoning-model weights and six smaller derivatives trained from its outputs. Its own tests put R1 near OpenAI o1 on several reasoning benchmarks, while the technical report documented a training approach others could study and limitations beyond those benchmarks.","category":"model","secondaryCategories":["open-source"],"orgs":["DeepSeek"],"tier":1,"source":{"url":"https://deepseek.com/en/news/deepseek-r1/","title":"DeepSeek-R1 Release","publisher":"DeepSeek","type":"announcement","date":"2025-01-20"},"references":[{"url":"https://api-docs.deepseek.com/news/news250120/","title":"DeepSeek-R1 Release","publisher":"DeepSeek","type":"announcement","date":"2025-01-20","locator":"Original announcement address; retained as provenance after the news page moved to deepseek.com."},{"url":"https://arxiv.org/abs/2501.12948","title":"DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning","publisher":"arXiv","type":"paper","date":"2025-01-22"},{"url":"https://github.com/deepseek-ai/DeepSeek-R1/blob/fdf883c01427a35878dba3e43a8eca5c765badb8/README.md","title":"DeepSeek-R1","publisher":"DeepSeek","type":"docs","date":"2025-01-21"}]}
{"id":"2025-01-20-liang-wenfeng-li-qiang-meeting","date":"2025-01-20","precision":"day","title":"DeepSeek founder Liang Wenfeng attends a meeting with China’s premier","summary":"Liang Wenfeng joined Premier Li Qiang’s January 20 symposium with business and technology leaders. The meeting drew attention to DeepSeek on the day R1 was released.","category":"culture","orgs":["DeepSeek"],"tier":3,"source":{"url":"https://www.scmp.com/tech/policy/article/3295662/beijing-meeting-puts-spotlight-chinas-new-face-ai-deepseek-founder-liang-wenfeng","title":"Beijing meeting puts spotlight on China's new face of AI, DeepSeek founder Liang Wenfeng","publisher":"South China Morning Post","type":"news","date":"2025-01-21"}}
{"id":"2025-01-21-stargate-project","date":"2025-01-21","precision":"day","title":"Stargate commits $500 billion to data centers","summary":"OpenAI, Oracle, SoftBank and MGX announced Stargate, a joint venture to develop AI data centers in the United States, with a stated investment target of up to $500 billion over four years.","category":"company","orgs":["OpenAI","Oracle","SoftBank","MGX"],"tier":2,"source":{"url":"https://openai.com/index/announcing-the-stargate-project/","title":"Announcing The Stargate Project","publisher":"OpenAI","type":"announcement","date":"2025-01-21"}}
{"id":"2025-01-22-moonshot-kimi-k1-5","date":"2025-01-22","precision":"day","title":"Moonshot AI ships Kimi k1.5, an o1-class reasoning model","summary":"Moonshot AI released Kimi k1.5, trained with reinforcement learning over context windows up to 128k tokens, and reported math and coding scores matching OpenAI's o1 without Monte Carlo tree search or a separate reward model. Code and a technical report shipped alongside it.","category":"model","orgs":["Moonshot AI"],"tier":3,"source":{"url":"https://arxiv.org/abs/2501.12599","title":"Kimi k1.5: Scaling Reinforcement Learning with LLMs","publisher":"arXiv","type":"paper","date":"2025-01-22"},"references":[{"url":"https://github.com/MoonshotAI/Kimi-k1.5","title":"Kimi-k1.5","publisher":"Moonshot AI","type":"docs"}]}
{"id":"2025-01-23-executive-order-14179","date":"2025-01-23","precision":"day","title":"An executive order trades the safety framework for an action plan","summary":"Executive Order 14179 revoked the 2023 order on AI safety three days into the new administration and gave the White House 180 days to produce a national action plan. Federal policy went from managing risk to removing barriers in a single document.","category":"policy","orgs":["White House"],"tier":2,"source":{"url":"https://www.whitehouse.gov/presidential-actions/2025/01/removing-barriers-to-american-leadership-in-artificial-intelligence/","title":"Removing Barriers to American Leadership in Artificial Intelligence","publisher":"The White House","type":"press-release","date":"2025-01-23"},"references":[{"url":"https://www.federalregister.gov/documents/2025/01/31/2025-02172/removing-barriers-to-american-leadership-in-artificial-intelligence","title":"Removing Barriers to American Leadership in Artificial Intelligence","publisher":"Federal Register","type":"docs","date":"2025-01-31","locator":"Executive Order 14179, signed January 23, 2025"}]}
{"id":"2025-01-23-openai-operator","date":"2025-01-23","precision":"day","title":"Operator drives the browser instead of calling an API","summary":"OpenAI launched Operator as a research preview for ChatGPT Pro subscribers. It could navigate websites to complete tasks, and OpenAI reported a 38 percent OSWorld score against a 72 percent human baseline.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://openai.com/index/introducing-operator/","title":"Introducing Operator","publisher":"OpenAI","type":"announcement","date":"2025-01-23"},"references":[{"url":"https://techcrunch.com/2025/01/23/openai-launches-operator-an-ai-agent-that-performs-tasks-autonomously/","title":"OpenAI launches Operator, an AI agent that performs tasks autonomously","publisher":"TechCrunch","type":"news","date":"2025-01-23"}]}
{"id":"2025-01-24-meta-60-65-billion-ai-capex","date":"2025-01-24","precision":"day","title":"Zuckerberg commits Meta to $60-65 billion in AI capex for 2025","summary":"Mark Zuckerberg said Meta would spend $60 billion to $65 billion on capital expenditures in 2025, mostly data centers and chips for AI, a roughly two-thirds increase over 2024. He said the buildout was meant to make Meta AI the most-used assistant of the year.","category":"deal","orgs":["Meta"],"tier":2,"source":{"url":"https://fortune.com/2025/01/24/mark-zuckerberg-ai-engineer-capex-spend/","title":"Mark Zuckerberg is building an “AI engineer” at Meta, ramping up spending to $65 billion","publisher":"Fortune","type":"news","date":"2025-01-24"}}
{"id":"2025-01-27-deepseek-janus-pro","date":"2025-01-27","precision":"day","title":"DeepSeek releases Janus-Pro, an open text-to-image model","summary":"DeepSeek open sourced Janus-Pro, a unified model for image understanding and text-to-image generation in 1B and 7B sizes, a week after R1 upended the reasoning-model market. It claimed better prompt following and more stable generation than the original Janus.","category":"model","secondaryCategories":["open-source"],"orgs":["DeepSeek"],"tier":3,"source":{"url":"https://github.com/deepseek-ai/Janus","title":"Janus-Pro","publisher":"DeepSeek","type":"docs","date":"2025-01-27"}}
{"id":"2025-01-27-deepseek-selloff","date":"2025-01-27","precision":"day","title":"Nvidia loses nearly $600 billion in market value after DeepSeek’s R1 release","summary":"Nvidia lost nearly $600 billion in market value on January 27 as investors questioned demand for costly AI computing after DeepSeek released R1.","category":"culture","orgs":["NVIDIA","DeepSeek"],"tier":2,"source":{"url":"https://www.cnbc.com/2025/01/27/nvidia-sheds-almost-600-billion-in-market-cap-biggest-drop-ever.html","title":"Nvidia sheds almost $600 billion in market cap, biggest one-day loss in U.S. history","publisher":"CNBC","type":"news","date":"2025-01-27"}}
{"id":"2025-01-28-alibaba-qwen2-5-max","date":"2025-01-28","precision":"day","title":"Alibaba launches Qwen2.5-Max, its flagship model, against DeepSeek","summary":"Alibaba released Qwen2.5-Max, a mixture-of-experts model pretrained on more than 20 trillion tokens, and published benchmark comparisons claiming wins over DeepSeek V3 on Arena-Hard, LiveBench and GPQA-Diamond. It shipped through Qwen Chat and Alibaba Cloud's API rather than as open weights.","category":"model","orgs":["Alibaba"],"tier":3,"source":{"url":"https://qwenlm.github.io/blog/qwen2.5-max/","title":"Qwen2.5-Max: Exploring the Intelligence of Large-scale MoE Model","publisher":"Qwen","type":"announcement","date":"2025-01-28"}}
{"id":"2025-01-28-openai-chatgpt-gov","date":"2025-01-28","precision":"day","title":"OpenAI launches ChatGPT Gov for US government agencies","summary":"OpenAI introduced ChatGPT Gov, a version of ChatGPT that agencies can deploy on Microsoft Azure's commercial or government clouds to handle sensitive but unclassified data. OpenAI said more than 90,000 government users had already been using ChatGPT Enterprise tools since 2024.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://techcrunch.com/2025/01/28/openai-launches-chatgpt-plan-for-u-s-government-agencies/","title":"OpenAI launches ChatGPT plan for US government agencies","publisher":"TechCrunch","type":"news","date":"2025-01-28"},"references":[{"url":"https://openai.com/global-affairs/introducing-chatgpt-gov/","title":"Introducing ChatGPT Gov","publisher":"OpenAI","type":"announcement","date":"2025-01-28"}]}
{"id":"2025-01-29-microsoft-openai-deepseek-distillation-probe","date":"2025-01-29","precision":"day","title":"Microsoft and OpenAI investigate DeepSeek over possible distillation","summary":"Microsoft's security team flagged accounts linked to DeepSeek pulling unusually large volumes of data through OpenAI's API in late 2024, and the companies opened a joint investigation into whether the outputs were used to train a rival model. OpenAI's terms bar using its outputs to build competing systems.","category":"culture","orgs":["Microsoft","OpenAI","DeepSeek"],"tier":3,"source":{"url":"https://www.tomshardware.com/tech-industry/artificial-intelligence/microsoft-and-open-ai-investigate-whether-deepseek-illicitly-obtained-data-from-chatgpt","title":"Microsoft and OpenAI investigate whether DeepSeek illicitly obtained data from ChatGPT","publisher":"Tom's Hardware","type":"news","date":"2025-01-29"}}
{"id":"2025-01-29-wiz-deepseek-exposed-database","date":"2025-01-29","precision":"day","title":"Wiz finds DeepSeek's database open on the internet, no password needed","summary":"Security researchers at Wiz found a publicly accessible DeepSeek ClickHouse database exposing chat histories, API keys and backend logs with no authentication required. DeepSeek secured the database after Wiz disclosed it.","category":"culture","orgs":["DeepSeek","Wiz"],"tier":3,"source":{"url":"https://www.wiz.io/blog/wiz-research-uncovers-exposed-deepseek-database-leak","title":"Wiz Research Uncovers Exposed DeepSeek Database Leaking Sensitive Information, Including Chat History","publisher":"Wiz","type":"report","date":"2025-01-29"}}
{"id":"2025-01-30-italy-blocks-deepseek","date":"2025-01-30","precision":"day","title":"Italy's data protection authority orders an emergency block on DeepSeek","summary":"The Garante ordered Hangzhou DeepSeek and Beijing DeepSeek to stop processing Italian users' data with immediate effect, calling the companies' answers about their data practices inadequate, and opened a formal investigation. It was the first national block of the chatbot.","category":"policy","orgs":["DeepSeek","Garante per la protezione dei dati personali"],"tier":3,"source":{"url":"https://www.garanteprivacy.it/garante/doc.jsp?ID=10097450","title":"Artificial Intelligence: The Italian Data Protection Authority blocks DeepSeek","publisher":"Garante per la protezione dei dati personali","type":"press-release","date":"2025-01-30"}}
{"id":"2025-01-30-mistral-small-3","date":"2025-01-30","precision":"day","title":"Mistral releases Small 3, a 24B open model tuned for speed","summary":"Mistral AI released Small 3, a 24 billion parameter model under Apache 2.0 that it said matched larger models like Llama 3.3 70B while running more than three times faster on comparable hardware. It reported 81 percent accuracy on MMLU at 150 tokens per second.","category":"model","secondaryCategories":["open-source"],"orgs":["Mistral AI"],"tier":3,"source":{"url":"https://mistral.ai/news/mistral-small-3/","title":"Mistral Small 3","publisher":"Mistral AI","type":"announcement","date":"2025-01-30"}}
{"id":"2025-01-31-openai-o3-mini","date":"2025-01-31","precision":"day","title":"OpenAI ships o3-mini, a cheaper reasoning model with a free tier","summary":"OpenAI released o3-mini, a smaller reasoning model priced 63 percent below o1-mini at $0.55 per million input tokens and $4.40 per million output tokens, competitive with DeepSeek R1's API pricing. Free ChatGPT users got limited access through a new Reason button, while paid users could pick a reasoning effort level directly.","category":"model","orgs":["OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2025/01/31/openai-launches-o3-mini-its-latest-reasoning-model/","title":"OpenAI launches o3-mini, its latest 'reasoning' model","publisher":"TechCrunch","type":"news","date":"2025-01-31"},"references":[{"url":"https://openai.com/index/openai-o3-mini/","title":"OpenAI o3-mini","publisher":"OpenAI","type":"announcement","date":"2025-01-31"}]}
{"id":"2025-02-02-openai-deep-research","date":"2025-02-02","precision":"day","title":"OpenAI launches Deep Research, an agent that browses and writes reports","summary":"A version of the o3 reasoning model that browses the web, reads files and synthesizes findings into a cited report over several minutes. Launched to $200 a month ChatGPT Pro subscribers first, capped at 100 queries a month.","category":"product","orgs":["OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2025/02/02/openai-unveils-a-new-chatgpt-agent-for-deep-research/","title":"OpenAI unveils a new ChatGPT agent for 'deep research'","publisher":"TechCrunch","type":"news","date":"2025-02-02"},"references":[{"url":"https://openai.com/index/introducing-deep-research/","title":"Introducing deep research","publisher":"OpenAI","type":"announcement","date":"2025-02-02"}]}
{"id":"2025-02-03-constitutional-classifiers","date":"2025-02-03","precision":"day","title":"Anthropic challenges the public to jailbreak its new safety filter","summary":"Constitutional Classifiers, trained on synthetic data to block jailbreaks with few false refusals, held through 3,000 hours of private red teaming. A wider public bounty running through February 10 found one participant who passed all eight challenge levels.","category":"research","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/constitutional-classifiers","title":"Constitutional Classifiers: Defending Against Universal Jailbreaks","publisher":"Anthropic","type":"announcement","date":"2025-02-03"}}
{"id":"2025-02-05-gemini-2-family-ga","date":"2025-02-05","precision":"day","title":"Google makes the Gemini 2.0 model family generally available","summary":"Gemini 2.0 Flash reached general availability with a million-token context window, alongside an experimental Gemini 2.0 Pro tuned for coding and a cost-efficient Flash-Lite in public preview.","category":"model","orgs":["Google"],"tier":2,"source":{"url":"https://blog.google/technology/google-deepmind/gemini-model-updates-february-2025/","title":"Gemini 2.0 is now available to everyone","publisher":"Google","type":"announcement","date":"2025-02-05"}}
{"id":"2025-02-06-mistral-le-chat-relaunch","date":"2025-02-06","precision":"day","title":"Mistral relaunches Le Chat with search, image generation and OCR","summary":"A rebuilt consumer app adding web search, document understanding, a code interpreter and Flux-powered image generation, with responses delivered at roughly 1,000 words per second. Free on iOS, Android and the web, with paid tiers for heavier use.","category":"product","orgs":["Mistral AI"],"tier":3,"source":{"url":"https://mistral.ai/news/all-new-le-chat","title":"The all new le Chat: Your AI assistant for life and work","publisher":"Mistral AI","type":"announcement","date":"2025-02-06"}}
{"id":"2025-02-10-france-ai-investment-pledge","date":"2025-02-10","precision":"day","title":"France pledges 109 billion euros in AI investment at the Paris summit","summary":"President Emmanuel Macron announced public and private commitments worth 109 billion euros for AI data centers and infrastructure, framed as France's answer to the United States' Stargate project.","category":"deal","orgs":["France"],"tier":3,"source":{"url":"https://www.euronews.com/next/2025/02/10/world-leaders-and-tech-giants-converge-in-paris-for-ai-summit","title":"World leaders and tech giants converge in Paris for AI summit","publisher":"Euronews","type":"news","date":"2025-02-10"}}
{"id":"2025-02-10-musk-bid-openai-nonprofit","date":"2025-02-10","precision":"day","title":"Musk-led group offers $97.4 billion for OpenAI's nonprofit","summary":"A consortium led by Elon Musk, including his own xAI, submitted an unsolicited $97.4 billion offer for the assets of OpenAI's nonprofit parent. Sam Altman mocked the bid on X the same day, and OpenAI's board formally rejected it days later.","category":"culture","orgs":["OpenAI","xAI"],"tier":2,"source":{"url":"https://techcrunch.com/2025/02/10/elon-musk-led-team-submits-97-4b-bid-for-openai/","title":"Elon Musk-led team submits $97.4B bid for OpenAI","publisher":"TechCrunch","type":"news","date":"2025-02-10"},"references":[{"url":"https://techcrunch.com/2025/02/11/how-musks-97-4b-bid-could-gum-up-openais-for-profit-conversion/","title":"How Musk's $97.4B bid could gum up OpenAI's for-profit conversion","publisher":"TechCrunch","type":"analysis","date":"2025-02-11"}]}
{"id":"2025-02-11-paris-summit-declaration","date":"2025-02-11","precision":"day","title":"US and UK refuse to sign the Paris AI Action Summit declaration","summary":"Sixty-one countries, including China and India, signed a declaration on inclusive and trustworthy AI at the close of the Paris summit. JD Vance used the summit stage to warn against heavy AI regulation, and the US and UK declined to sign.","category":"policy","orgs":["White House"],"tier":2,"source":{"url":"https://techcrunch.com/2025/02/11/as-us-and-uk-refuse-to-sign-ai-action-summit-statement-countries-fail-to-agree-on-the-basics/","title":"As US and UK refuse to sign the Paris AI Action Summit statement, countries fail to agree on the basics","publisher":"TechCrunch","type":"news","date":"2025-02-11"},"references":[{"url":"https://www.euronews.com/next/2025/02/11/heres-what-has-been-announced-at-the-ai-action-summit","title":"Here's what has been announced at the AI Action Summit","publisher":"Euronews","type":"news","date":"2025-02-11"}]}
{"id":"2025-02-11-thomson-reuters-ross-fair-use-ruling","date":"2025-02-11","precision":"day","title":"A federal judge rules AI training on Westlaw headnotes was not fair use","summary":"Judge Stephanos Bibas found that ROSS Intelligence's use of Thomson Reuters's Westlaw headnotes to train a legal search tool infringed copyright and was not protected by fair use, the first US merits ruling on AI training data and copyright.","category":"culture","orgs":["Thomson Reuters","ROSS Intelligence"],"tier":2,"source":{"url":"https://www.ded.uscourts.gov/sites/ded/files/opinions/20-613_5.pdf","title":"Thomson Reuters Enterprise Centre GmbH v. ROSS Intelligence Inc., Memorandum Opinion","publisher":"U.S. District Court for the District of Delaware","type":"report","date":"2025-02-11"},"references":[{"url":"https://www.goodwinlaw.com/en/insights/publications/2025/02/alerts-practices-ip-lit-court-rejects-fair-use-defense-in-ai-copyright-case","title":"Court Rejects Fair Use Defense in AI Copyright Case","publisher":"Goodwin","type":"analysis","date":"2025-02-12"}]}
{"id":"2025-02-14-uk-ai-security-institute-rename","date":"2025-02-14","precision":"day","title":"The UK renames its AI Safety Institute the AI Security Institute","summary":"The rebrand narrows the institute's remit to serious security risks such as bioweapons, cyberattacks and fraud, dropping bias and free speech from its scope. It coincided with a new UK partnership with Anthropic on public services.","category":"policy","orgs":["UK AI Security Institute"],"tier":3,"source":{"url":"https://www.gov.uk/government/news/tackling-ai-security-risks-to-unleash-growth-and-deliver-plan-for-change","title":"Tackling AI security risks to unleash growth and deliver Plan for Change","publisher":"GOV.UK","type":"announcement","date":"2025-02-14"}}
{"id":"2025-02-17-mistral-saba","date":"2025-02-17","precision":"day","title":"Mistral ships Saba, a regional model for Arabic and South Asian languages","summary":"A 24 billion parameter model trained on curated Middle Eastern and South Asian data, including strong Arabic and Tamil support. Mistral pitched it as faster and cheaper than general-purpose models several times its size, with an option to deploy entirely on premises.","category":"model","orgs":["Mistral AI"],"tier":3,"source":{"url":"https://mistral.ai/news/mistral-saba","title":"Mistral Saba","publisher":"Mistral AI","type":"announcement","date":"2025-02-17"}}
{"id":"2025-02-17-south-korea-deepseek-suspension","date":"2025-02-17","precision":"day","title":"South Korea pulls DeepSeek from its app stores over data privacy","summary":"The Personal Information Protection Commission removed DeepSeek's app from Apple and Google's Korean storefronts after the company failed to answer questions about how it handles user data, part of a wider run of government bans following DeepSeek's January debut.","category":"policy","orgs":["DeepSeek"],"tier":3,"source":{"url":"https://www.aljazeera.com/news/2025/2/17/south-korea-removes-deepseek-from-app-stores-pending-privacy-review","title":"South Korea removes DeepSeek from app stores pending privacy review","publisher":"Al Jazeera","type":"news","date":"2025-02-17"}}
{"id":"2025-02-19-ai-co-scientist","date":"2025-02-19","precision":"day","title":"Google gives the scientific method a multi-agent system","summary":"An AI co-scientist on Gemini 2.0, with separate agents to generate, debate, rank, and evolve hypotheses. Lab work backed several of its suggestions, including a gene transfer mechanism another group had already confirmed and not yet published.","category":"research","orgs":["Google"],"tier":3,"source":{"url":"https://research.google/blog/accelerating-scientific-breakthroughs-with-an-ai-co-scientist/","title":"Accelerating scientific breakthroughs with an AI co-scientist","publisher":"Google Research","type":"announcement","date":"2025-02-19"}}
{"id":"2025-02-19-grok-3","date":"2025-02-19","precision":"day","title":"xAI launches Grok 3, trained on ten times the compute of Grok 2","summary":"Grok 3 topped the LMArena leaderboard on release and shipped with DeepSearch, an agent for searching across the web, plus a reasoning mode that scored 93 percent on the 2025 AIME. It rolled out first to X Premium subscribers.","category":"model","orgs":["xAI"],"tier":2,"source":{"url":"https://x.ai/news/grok-3","title":"Grok 3 Beta — The Age of Reasoning Agents","publisher":"xAI","type":"announcement","date":"2025-02-19"}}
{"id":"2025-02-20-figure-helix","date":"2025-02-20","precision":"day","title":"Figure's Helix model gives its humanoid robot fast, full-body control","summary":"A vision-language-action model that runs the robot's fingers, wrists, torso and head at 200Hz from a single network, letting two robots coordinate on a shared task and handle novel objects with no task-specific tuning. Trained on about 500 hours of data, far less than prior systems.","category":"research","orgs":["Figure"],"tier":3,"source":{"url":"https://www.figure.ai/news/helix","title":"Helix: A Vision-Language-Action Model for Generalist Humanoid Control","publisher":"Figure","type":"announcement","date":"2025-02-20"}}
{"id":"2025-02-21-deepseek-open-source-week","date":"2025-02-21","endDate":"2025-02-28","precision":"day","title":"DeepSeek open sources the infrastructure behind V3 and R1","summary":"Five repositories in five days, plus a bonus release: an MoE decoding kernel, a communication library, an FP8 GEMM library, pipeline-parallelism tools and a distributed file system, all described as production code. A closing report detailed the economics of running the models at scale.","category":"open-source","orgs":["DeepSeek"],"tier":2,"source":{"url":"https://github.com/deepseek-ai/open-infra-index","title":"DeepSeek Open Infra: Open-Sourcing 5 AI Repos in 5 Days","publisher":"DeepSeek","type":"docs","date":"2025-02-21"},"references":[{"url":"https://github.com/deepseek-ai/FlashMLA","title":"FlashMLA","publisher":"DeepSeek","type":"docs","date":"2025-02-24"}]}
{"id":"2025-02-24-alibaba-ai-cloud-investment","date":"2025-02-24","precision":"day","title":"Alibaba commits $53 billion to AI and cloud infrastructure","summary":"A three-year spending plan Alibaba said exceeds its total AI and cloud investment over the prior decade, announced alongside six straight quarters of triple-digit growth in AI product revenue. CEO Eddie Wu named artificial general intelligence the company's primary long-term goal.","category":"deal","orgs":["Alibaba"],"tier":3,"source":{"url":"https://www.alibabacloud.com/blog/alibaba-to-invest-rmb380-billion-in-ai-and-cloud-infrastructure-over-next-three-years_602007","title":"Alibaba to Invest RMB380 billion in AI and Cloud Infrastructure Over Next Three Years","publisher":"Alibaba Cloud","type":"announcement","date":"2025-02-24"}}
{"id":"2025-02-24-apple-500-billion-us-investment","date":"2025-02-24","precision":"day","title":"Apple commits $500 billion to US manufacturing, including AI servers","summary":"The pledge includes a new Houston factory building servers for Apple Intelligence and Private Cloud Compute, plus a doubled Advanced Manufacturing Fund and 20,000 planned US hires. Apple called it the largest domestic investment in its history.","category":"deal","orgs":["Apple"],"tier":2,"source":{"url":"https://www.apple.com/newsroom/2025/02/apple-will-spend-more-than-500-billion-usd-in-the-us-over-the-next-four-years/","title":"Apple will spend more than $500 billion USD in the US over the next four years","publisher":"Apple","type":"press-release","date":"2025-02-24"}}
{"id":"2025-02-24-claude-3-7-sonnet","date":"2025-02-24","precision":"day","title":"Anthropic releases Claude 3.7 Sonnet with extended thinking","summary":"Anthropic released Claude 3.7 Sonnet with standard and extended-thinking modes, API control over thinking tokens, and access across Claude plans and major cloud platforms.","category":"model","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/news/claude-3-7-sonnet","title":"Claude 3.7 Sonnet and Claude Code","publisher":"Anthropic","type":"announcement","date":"2025-02-24"}}
{"id":"2025-02-24-claude-code","date":"2025-02-24","precision":"day","title":"Claude Code puts an agent in the terminal","summary":"Anthropic launched Claude Code alongside Claude 3.7 Sonnet as a limited research preview. The command-line tool could read a repository, propose file edits and run approved shell commands.","category":"product","orgs":["Anthropic"],"tier":1,"source":{"url":"https://www.anthropic.com/news/claude-3-7-sonnet","title":"Claude 3.7 Sonnet and Claude Code","publisher":"Anthropic","type":"announcement","date":"2025-02-24"}}
{"id":"2025-02-24-mongodb-acquires-voyage-ai","date":"2025-02-24","precision":"day","title":"MongoDB acquires the AI retrieval startup Voyage AI","summary":"Voyage AI's embedding and reranking models, built to reduce hallucinations in retrieval-augmented systems, will fold into MongoDB's database rather than remain a standalone product. Voyage's existing API and cloud marketplace listings continue for now.","category":"acquisition","orgs":["MongoDB","Voyage AI"],"tier":3,"source":{"url":"https://www.prnewswire.com/news-releases/mongodb-announces-acquisition-of-voyage-ai-to-enable-organizations-to-build-trustworthy-ai-applications-302382979.html","title":"MongoDB Announces Acquisition of Voyage AI to Enable Organizations to Build Trustworthy AI Applications","publisher":"MongoDB","type":"press-release","date":"2025-02-24"}}
{"id":"2025-02-26-amazon-alexa-plus","date":"2025-02-26","precision":"day","title":"Amazon relaunches Alexa as a generative AI subscription assistant","summary":"Alexa+ rebuilds the assistant around a large language model, adding conversational memory and task-completing 'experts' for things like restaurant bookings. It costs $19.99 a month, free for Prime members, and began rolling out in the US.","category":"product","orgs":["Amazon"],"tier":2,"source":{"url":"https://www.aboutamazon.com/news/devices/new-alexa-generative-artificial-intelligence","title":"Introducing Alexa+, the next generation of Alexa","publisher":"Amazon","type":"announcement","date":"2025-02-26"}}
{"id":"2025-02-26-inception-labs-mercury","date":"2025-02-26","precision":"day","title":"Inception Labs launches Mercury, a diffusion-based language model","summary":"A Stanford-founded startup built a language model that denoises text in parallel rather than generating it token by token, claiming over 1,000 tokens per second on commodity GPUs. Mercury Coder, its code-focused variant, is the first product.","category":"model","orgs":["Inception Labs"],"tier":3,"source":{"url":"https://techcrunch.com/2025/02/26/inception-emerges-from-stealth-with-a-new-type-of-ai-model/","title":"Inception emerges from stealth with a new type of AI model","publisher":"TechCrunch","type":"news","date":"2025-02-26"}}
{"id":"2025-02-26-nvidia-q4-fy25-earnings","date":"2025-02-26","precision":"day","title":"Nvidia posts record data center revenue in Blackwell's first full quarter","summary":"Fourth-quarter data center revenue reached $35.6 billion, up 93 percent year over year, as Blackwell chips ramped into full production. Jensen Huang said demand for the new architecture was outstripping supply.","category":"hardware","orgs":["NVIDIA"],"tier":3,"source":{"url":"https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-fourth-quarter-and-fiscal-2025","title":"NVIDIA Announces Financial Results for Fourth Quarter and Fiscal 2025","publisher":"NVIDIA","type":"press-release","date":"2025-02-26"}}
{"id":"2025-02-26-phi-4-multimodal","date":"2025-02-26","precision":"day","title":"Microsoft ships Phi-4-multimodal and Phi-4-mini for edge devices","summary":"Phi-4-multimodal, Microsoft's first model to combine speech, vision and text in one small network, topped the HuggingFace OpenASR leaderboard. The text-only Phi-4-mini targets phones and other compute-constrained hardware.","category":"model","orgs":["Microsoft"],"tier":3,"source":{"url":"https://azure.microsoft.com/en-us/blog/empowering-innovation-the-next-generation-of-the-phi-family/","title":"Empowering innovation: The next generation of the Phi family","publisher":"Microsoft Azure","type":"announcement","date":"2025-02-26"}}
{"id":"2025-02-27-gpt-4-5-preview","date":"2025-02-27","precision":"day","title":"OpenAI unveils GPT-4.5, its largest model, at a steep price","summary":"Code-named Orion, GPT-4.5 launched first to $200-a-month ChatGPT Pro subscribers and developers, priced at $75 per million input tokens. OpenAI's own materials briefly said it was not a frontier model, before that language was removed hours after release.","category":"model","orgs":["OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2025/02/27/openai-unveils-gpt-4-5-orion-its-largest-ai-model-yet/","title":"OpenAI unveils GPT-4.5 'Orion,' its largest AI model yet","publisher":"TechCrunch","type":"news","date":"2025-02-27"}}
{"id":"2025-03-manus-viral-launch","date":"2025-03","precision":"month","title":"Manus, a Chinese AI agent, goes viral in an invite-only preview","summary":"Butterfly Effect's agent routed tasks to Claude and Qwen rather than models of its own, and a demo video sent its Discord server past 138,000 members within days, with invite codes reselling for thousands of dollars.","category":"product","orgs":["Manus"],"tier":3,"source":{"url":"https://techcrunch.com/2025/03/09/manus-probably-isnt-chinas-second-deepseek-moment/","title":"Manus probably isn't China's second 'DeepSeek moment'","publisher":"TechCrunch","type":"news","date":"2025-03-09"},"references":[{"url":"https://www.scmp.com/tech/tech-trends/article/3301864/chinese-ai-agent-manus-transcends-chatbots-founder-start-butterfly-effect-says","title":"Chinese AI agent Manus transcends chatbots, founder of start-up Butterfly Effect says","publisher":"South China Morning Post","type":"news","date":"2025-03-11"}]}
{"id":"2025-03-03-anthropic-series-e","date":"2025-03-03","precision":"day","title":"Anthropic raises $3.5 billion Series E at a $61.5 billion valuation","summary":"Lightspeed Venture Partners led the round, with Fidelity, General Catalyst, Jane Street and Salesforce Ventures among the backers, nearly tripling Anthropic's valuation from a year earlier.","category":"funding","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/news/anthropic-raises-series-e-at-usd61-5b-post-money-valuation","title":"Anthropic raises Series E at $61.5B post-money valuation","publisher":"Anthropic","type":"announcement","date":"2025-03-03"},"references":[{"url":"https://techcrunch.com/2025/03/03/anthropic-raises-3-5b-to-fuel-its-ai-ambitions/","title":"Anthropic raises $3.5B to fuel its AI ambitions","publisher":"TechCrunch","type":"news","date":"2025-03-03"}]}
{"id":"2025-03-05-gpt-4-5-chatgpt-plus","date":"2025-03-05","precision":"day","title":"GPT-4.5 expands from ChatGPT Pro to Plus subscribers","summary":"OpenAI's largest and most expensive model to date reached wider ChatGPT access a week after its Pro-only debut, even as it trailed newer reasoning models from DeepSeek and Anthropic on several benchmarks.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://techcrunch.com/2025/03/05/openais-gpt-4-5-ai-model-comes-to-more-chatgpt-users/","title":"OpenAI's GPT-4.5 AI model comes to more ChatGPT users","publisher":"TechCrunch","type":"news","date":"2025-03-05"},"references":[{"url":"https://techcrunch.com/2025/02/27/openais-gpt-4-5-is-better-at-convincing-other-ai-to-give-it-money/","title":"OpenAI's GPT-4.5 is better at convincing other AIs to give it money","publisher":"TechCrunch","type":"news","date":"2025-02-27"}]}
{"id":"2025-03-06-mistral-ocr","date":"2025-03-06","precision":"day","title":"Mistral ships an OCR API that turns PDFs into AI-ready Markdown","summary":"The model parses tables, equations and multi-column layouts at about 2,000 pages a minute, priced at 1,000 pages per dollar, and claims higher accuracy than GPT-4o and Gemini on Mistral's own benchmark.","category":"product","orgs":["Mistral AI"],"tier":3,"source":{"url":"https://mistral.ai/news/mistral-ocr","title":"Mistral OCR","publisher":"Mistral AI","type":"announcement","date":"2025-03-06"},"references":[{"url":"https://techcrunch.com/2025/03/06/mistrals-new-ocr-api-turns-any-pdf-document-into-an-ai-ready-markdown-file/","title":"Mistral's new OCR API turns any PDF document into an AI-ready Markdown file","publisher":"TechCrunch","type":"news","date":"2025-03-06"}]}
{"id":"2025-03-12-gemini-2-0-flash-image-gen","date":"2025-03-12","precision":"day","title":"Gemini 2.0 Flash gets experimental native image generation","summary":"Google opened multi-turn conversational image creation and editing to developers in AI Studio through the gemini-2.0-flash-exp model, following a limited trusted-tester rollout in December.","category":"product","orgs":["Google"],"tier":3,"source":{"url":"https://developers.googleblog.com/en/experiment-with-gemini-20-flash-native-image-generation/","title":"Experiment with Gemini 2.0 Flash native image generation","publisher":"Google","type":"announcement","date":"2025-03-12"},"references":[{"url":"https://techcrunch.com/2025/03/25/chatgpts-image-generation-feature-gets-an-upgrade/","title":"ChatGPT's image-generation feature gets an upgrade","publisher":"TechCrunch","type":"news","date":"2025-03-25"}]}
{"id":"2025-03-12-gemini-robotics","date":"2025-03-12","precision":"day","title":"Gemini Robotics brings a vision-language-action model to real robots","summary":"Built on Gemini 2.0 with physical actions as a new output, the model handles novel objects and multi-step manipulation like origami folding, with Apptronik, Boston Dynamics and Agility Robotics as early testers.","category":"model","orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://deepmind.google/discover/blog/gemini-robotics-brings-ai-into-the-physical-world/","title":"Gemini Robotics brings AI into the physical world","publisher":"Google DeepMind","type":"announcement","date":"2025-03-12"},"references":[{"url":"https://techcrunch.com/2025/03/12/google-deepmind-unveils-new-ai-models-for-controlling-robots/","title":"Google DeepMind unveils new AI models for controlling robots","publisher":"TechCrunch","type":"news","date":"2025-03-12"}]}
{"id":"2025-03-12-gemma-3","date":"2025-03-12","precision":"day","title":"Gemma 3 fits a capable open model on one GPU","summary":"Four sizes from 1B to 27B, vision above the smallest, and up to 128,000 tokens of context, with Google claiming the 27B beat models many times its size. The hardware requirement was the interesting number, not the benchmark.","category":"model","secondaryCategories":["open-source"],"orgs":["Google"],"tier":3,"source":{"url":"https://blog.google/technology/developers/gemma-3/","title":"Introducing Gemma 3: The most capable model you can run on a single GPU or TPU","publisher":"Google","type":"announcement","date":"2025-03-12"}}
{"id":"2025-03-16-baidu-ernie-4-5-x1","date":"2025-03-16","precision":"day","title":"Baidu launches Ernie 4.5 and the reasoning-focused Ernie X1","summary":"Ernie 4.5 adds multimodal understanding Baidu pitches as reading memes and satire, while Ernie X1 claims DeepSeek R1-level reasoning at half the price, as Baidu works to keep pace with China's AI rivals.","category":"model","orgs":["Baidu"],"tier":3,"source":{"url":"https://techcrunch.com/2025/03/16/baidu-launches-two-new-versions-of-its-ai-model-ernie/","title":"Baidu launches two new versions of its AI model Ernie","publisher":"TechCrunch","type":"news","date":"2025-03-16"}}
{"id":"2025-03-17-amazon-ends-local-alexa-processing","date":"2025-03-17","precision":"day","title":"Amazon forces all Alexa voice processing to the cloud for Alexa+","summary":"Citing the compute demands of its new generative AI features, Amazon ended on-device processing on the three Echo models that supported it, removing the option to keep voice recordings off its servers starting March 28.","category":"product","orgs":["Amazon"],"tier":3,"source":{"url":"https://www.theregister.com/2025/03/17/amazon_kills_on_device_alexa/","title":"Amazon kills on-device Alexa privacy setting for Echo devices","publisher":"The Register","type":"news","date":"2025-03-17"}}
{"id":"2025-03-17-nvidia-gr00t-n1","date":"2025-03-17","precision":"day","title":"Nvidia releases Isaac GR00T N1, an open humanoid robot foundation model","summary":"Described by Nvidia as the first open foundation model for general humanoid reasoning, it turns camera images, robot state and text instructions into motor actions across different robot bodies, released in 1B and 2B parameter sizes.","category":"model","orgs":["NVIDIA"],"tier":3,"source":{"url":"https://huggingface.co/nvidia/GR00T-N1-2B","title":"nvidia/GR00T-N1-2B","publisher":"NVIDIA","type":"docs","date":"2025-03-17"}}
{"id":"2025-03-18-google-agrees-to-buy-wiz","date":"2025-03-18","precision":"day","title":"Google agrees to buy Wiz for $32 billion","summary":"The largest acquisition in Alphabet's history, all cash, for a cloud security company five years old. An earlier run at $23 billion had fallen apart the previous summer.","category":"acquisition","orgs":["Google","Wiz"],"tier":2,"source":{"url":"https://blog.google/company-news/inside-google/company-announcements/google-agreement-acquire-wiz/","title":"Google announces agreement to acquire Wiz","publisher":"Google","type":"announcement","date":"2025-03-18"},"references":[{"url":"https://techcrunch.com/2025/03/18/google-is-buying-wiz-for-32b-to-beef-up-in-cloud-security/","title":"Confirmed: Google buys Wiz for $32B to beef up in cloud security","publisher":"TechCrunch","type":"news","date":"2025-03-18"}]}
{"id":"2025-03-18-nvidia-gtc-vera-rubin","date":"2025-03-18","precision":"day","title":"Nvidia previews Vera Rubin GPUs and ships DGX Spark at GTC 2025","summary":"Jensen Huang's keynote promised roughly double Blackwell's inference speed from the next-generation Vera Rubin platform, alongside new DGX Spark and DGX Station desktop AI computers and a Boston quantum computing research center.","category":"hardware","orgs":["NVIDIA"],"tier":2,"source":{"url":"https://techcrunch.com/2025/03/20/gtc-felt-more-bullish-than-ever-but-nvidias-challenges-are-piling-up/","title":"GTC felt more bullish than ever, but Nvidia's challenges are piling up","publisher":"TechCrunch","type":"news","date":"2025-03-20"},"references":[{"url":"https://www.theregister.com/2025/03/29/nvidia_moores_law/","title":"Nvidia GPU roadmap confirms it: Moore's Law is dead and buried","publisher":"The Register","type":"analysis","date":"2025-03-29"}]}
{"id":"2025-03-20-apple-siri-ai-shakeup","date":"2025-03-20","precision":"day","title":"Apple puts its Vision Pro lead in charge of a stalled Siri","summary":"Mike Rockwell took over Siri from AI chief John Giannandrea after Tim Cook reportedly lost confidence in the group's ability to ship the overhauled assistant Apple had promoted alongside the iPhone 16.","category":"company","orgs":["Apple"],"tier":3,"source":{"url":"https://techcrunch.com/2025/03/20/apple-puts-vision-pro-exec-in-charge-of-siri-in-exec-shakeup/","title":"Apple puts Vision Pro exec in charge of Siri in exec shakeup","publisher":"TechCrunch","type":"news","date":"2025-03-20"}}
{"id":"2025-03-20-claude-web-search","date":"2025-03-20","precision":"day","title":"Claude gains the ability to search the web","summary":"Anthropic switched on live web search for paid Claude 3.7 Sonnet users in the United States, attaching citations to sourced claims ahead of a later expansion to free users and more countries.","category":"product","orgs":["Anthropic"],"tier":3,"source":{"url":"https://claude.com/blog/web-search","title":"Claude can now search the web","publisher":"Anthropic","type":"announcement","date":"2025-03-20"},"references":[{"url":"https://techcrunch.com/2025/03/20/anthropic-adds-web-search-to-its-claude-chatbot/","title":"Anthropic adds web search to its Claude chatbot","publisher":"TechCrunch","type":"news","date":"2025-03-20"}]}
{"id":"2025-03-20-softbank-acquires-ampere","date":"2025-03-20","precision":"day","title":"SoftBank agrees to buy chip designer Ampere Computing for $6.5 billion","summary":"The all-cash deal hands SoftBank a maker of energy-efficient Arm server processors as it builds out AI infrastructure alongside its Stargate commitments, with Carlyle and Oracle exiting as lead investors.","category":"acquisition","orgs":["SoftBank","Ampere Computing"],"tier":2,"source":{"url":"https://group.softbank/en/news/press/20250320","title":"SoftBank Group to Acquire Ampere Computing","publisher":"SoftBank Group","type":"press-release","date":"2025-03-20"},"references":[{"url":"https://techcrunch.com/2025/03/19/softbank-to-acquire-semiconductor-designer-ampere-in-6-5b-all-cash-deal/","title":"SoftBank to acquire semiconductor designer Ampere in $6.5B all-cash deal","publisher":"TechCrunch","type":"news","date":"2025-03-19"}]}
{"id":"2025-03-24-deepseek-v3-0324","date":"2025-03-24","precision":"day","title":"DeepSeek updates V3 and releases it under the MIT license","summary":"The refreshed checkpoint lifted reasoning scores such as AIME from 39.6 to 59.4 and fixed function-calling bugs, while switching from DeepSeek's custom license to MIT for unrestricted reuse.","category":"model","secondaryCategories":["open-source"],"orgs":["DeepSeek"],"tier":3,"source":{"url":"https://huggingface.co/deepseek-ai/DeepSeek-V3-0324","title":"deepseek-ai/DeepSeek-V3-0324","publisher":"DeepSeek","type":"docs","date":"2025-03-24"},"references":[{"url":"https://simonwillison.net/2025/Mar/24/deepseek/","title":"DeepSeek releases their latest DeepSeek-v3 model, now the best available open weights model","publisher":"Simon Willison","type":"analysis","date":"2025-03-24"}]}
{"id":"2025-03-24-qwen2-5-vl-32b","date":"2025-03-24","precision":"day","title":"Alibaba releases Qwen2.5-VL-32B, a smaller vision-language model","summary":"The 32-billion-parameter model, tuned with reinforcement learning for more human-aligned responses, beat Alibaba's own larger 72B model on math and image-understanding benchmarks under an Apache 2.0 license.","category":"model","secondaryCategories":["open-source"],"orgs":["Alibaba"],"tier":3,"source":{"url":"https://qwenlm.github.io/blog/qwen2.5-vl-32b/","title":"Qwen2.5-VL-32B: Smarter and Lighter","publisher":"Alibaba","type":"announcement","date":"2025-03-24"}}
{"id":"2025-03-25-gemini-2-5-pro","date":"2025-03-25","precision":"day","title":"Gemini 2.5 Pro puts Google back in the argument","summary":"Google released Gemini 2.5 Pro as an experimental model in AI Studio and reported leading results on several math and science benchmarks.","category":"model","orgs":["Google DeepMind"],"tier":2,"source":{"url":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","title":"Gemini 2.5: Our most intelligent AI model","publisher":"Google","type":"announcement","date":"2025-03-25"}}
{"id":"2025-03-25-gpt-4o-native-image-generation","date":"2025-03-25","precision":"day","title":"GPT-4o gets native image generation and sets off a Studio Ghibli trend","summary":"OpenAI replaced DALL-E 3 inside ChatGPT with GPT-4o's own image output, and users immediately flooded social feeds with Ghibli-style portraits, straining OpenAI's GPUs and reviving debate over training on copyrighted art styles.","category":"product","orgs":["OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2025/03/25/chatgpts-image-generation-feature-gets-an-upgrade/","title":"ChatGPT's image-generation feature gets an upgrade","publisher":"TechCrunch","type":"news","date":"2025-03-25"},"references":[{"url":"https://techcrunch.com/2025/03/31/openais-new-image-generator-is-now-available-to-all-users/","title":"OpenAI's new image generator is now available to all users","publisher":"TechCrunch","type":"news","date":"2025-03-31"},{"url":"https://openai.com/index/introducing-4o-image-generation/","title":"Introducing 4o Image Generation","publisher":"OpenAI","type":"announcement","date":"2025-03-25"}]}
{"id":"2025-03-27-anthropic-circuit-tracing","date":"2025-03-27","precision":"day","title":"Anthropic traces how Claude 'thinks' with a new interpretability method","summary":"An attribution-graph technique showed Claude planning ahead when writing rhymes, sharing concepts across languages, and sometimes fabricating reasoning to agree with a user rather than follow logic.","category":"research","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/research/tracing-thoughts-language-model","title":"Tracing the thoughts of a large language model","publisher":"Anthropic","type":"report","date":"2025-03-27"},"references":[{"url":"https://transformer-circuits.pub/2025/attribution-graphs/methods.html","title":"Circuit Tracing: Revealing Computational Graphs in Language Models","publisher":"Transformer Circuits Thread","type":"paper","date":"2025-03-27"}]}
{"id":"2025-03-28-coreweave-ipo","date":"2025-03-28","precision":"day","title":"CoreWeave goes public in the largest AI-related IPO to date","summary":"The Nvidia-backed cloud provider priced below its target range to raise $1.5 billion, a scaled-back debut that was still the biggest US tech IPO since 2021.","category":"funding","orgs":["CoreWeave"],"tier":2,"source":{"url":"https://techcrunch.com/2025/03/29/coreweave-co-founder-explains-how-a-closet-of-crypto-mining-gpus-led-to-a-1-5b-ipo/","title":"CoreWeave co-founder explains how a closet of crypto-mining GPUs led to a $1.5B IPO","publisher":"TechCrunch","type":"news","date":"2025-03-29"}}
{"id":"2025-03-28-xai-acquires-x","date":"2025-03-28","precision":"day","title":"xAI buys X in an all-stock deal","summary":"Musk sold one of his companies to another, valuing X at $33 billion and xAI at $80 billion. The training data, the distribution and the models now sat inside one entity.","category":"acquisition","orgs":["xAI","X"],"tier":2,"source":{"url":"https://www.cnbc.com/2025/03/28/elon-musk-says-xai-has-acquired-x-in-deal-that-values-social-media-site-at-33-billion.html","title":"Elon Musk says xAI has acquired X in deal that values social media site at $33 billion","publisher":"CNBC","type":"news","date":"2025-03-28"}}
{"id":"2025-03-31-openai-40b-round","date":"2025-03-31","precision":"day","title":"OpenAI closes a $40 billion round at a $300 billion valuation","summary":"SoftBank led the largest private tech funding round on record, with about $18 billion earmarked for the Stargate data center buildout, as OpenAI reported 500 million weekly ChatGPT users.","category":"funding","orgs":["OpenAI","SoftBank"],"tier":2,"source":{"url":"https://techcrunch.com/2025/03/31/openai-raises-40b-at-300b-post-money-valuation/","title":"OpenAI raises $40B at $300B post-money valuation","publisher":"TechCrunch","type":"news","date":"2025-03-31"},"references":[{"url":"https://techcrunch.com/2025/03/26/openai-will-reportedly-close-its-softbank-led-40-billion-round-soon/","title":"OpenAI will reportedly close its SoftBank-led $40B round soon","publisher":"TechCrunch","type":"news","date":"2025-03-26"}]}
{"id":"2025-03-31-runway-gen-4","date":"2025-03-31","precision":"day","title":"Runway launches Gen-4 for consistent AI video generation","summary":"The model keeps characters, objects and settings consistent across shots from a single reference image, aiming to make AI video usable for production work rather than one-off clips.","category":"model","orgs":["Runway"],"tier":3,"source":{"url":"https://techcrunch.com/2025/03/31/runway-releases-an-impressive-new-video-generating-ai-model/","title":"Runway releases an impressive new video-generating AI model","publisher":"TechCrunch","type":"news","date":"2025-03-31"},"references":[{"url":"https://runway.com/research/introducing-runway-gen-4","title":"Introducing Runway Gen-4","publisher":"Runway","type":"announcement","date":"2025-03-31"}]}
{"id":"2025-04-02-claude-for-education","date":"2025-04-02","precision":"day","title":"Anthropic launches Claude for Education","summary":"A version of Claude for universities with a Socratic Learning mode that guides students toward answers instead of giving them outright. Northeastern, the London School of Economics and Champlain College signed on for campus-wide access at launch.","category":"product","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/introducing-claude-for-education","title":"Introducing Claude for Education","publisher":"Anthropic","type":"announcement","date":"2025-04-02"}}
{"id":"2025-04-04-sec-gemini-v1","date":"2025-04-04","precision":"day","title":"Google announces Sec-Gemini v1, an experimental cybersecurity model","summary":"A Gemini variant built for security work, pairing Google's threat intelligence and vulnerability databases with reasoning. Aimed at tasks like incident root-cause analysis and threat summarization rather than general chat.","category":"model","orgs":["Google"],"tier":3,"source":{"url":"https://blog.google/security/google-launches-sec-gemini-v1-new/","title":"Google announces Sec-Gemini v1, a new experimental cybersecurity model","publisher":"Google","type":"announcement","date":"2025-04-04"}}
{"id":"2025-04-05-llama-4","date":"2025-04-05","precision":"day","title":"Meta releases Llama 4 Scout and Maverick","summary":"Two mixture of experts models with open weights, published on a Saturday to a reception Meta had not planned for. It was the last Llama release before the company rebuilt its AI organization around a new lab.","category":"model","secondaryCategories":["open-source"],"orgs":["Meta"],"tier":3,"source":{"url":"https://ai.meta.com/blog/llama-4-multimodal-intelligence/","title":"The Llama 4 herd: The beginning of a new era of natively multimodal AI innovation","publisher":"Meta","type":"announcement","date":"2025-04-05"},"references":[{"url":"https://techcrunch.com/2025/04/05/meta-releases-llama-4-a-new-crop-of-flagship-ai-models/","title":"Meta releases Llama 4, a new crop of flagship AI models","publisher":"TechCrunch","type":"news","date":"2025-04-05"}]}
{"id":"2025-04-08-llama-4-benchmark-controversy","date":"2025-04-08","precision":"day","title":"Meta accused of gaming the LM Arena leaderboard with Llama 4","summary":"Meta submitted an unreleased Llama 4 Maverick variant tuned for human preference voting to LM Arena, then shipped a plainer model to everyone else. LM Arena said Meta should have disclosed the substitution.","category":"culture","orgs":["Meta"],"tier":3,"source":{"url":"https://www.theregister.com/software/2025/04/08/meta-accused-of-llama-4-bait-n-switch-to-juice-lmarena-rank/1470166","title":"Meta accused of Llama 4 bait-and-switch to juice AI benchmark rank","publisher":"The Register","type":"news","date":"2025-04-08"},"references":[{"url":"https://techcrunch.com/2025/04/11/metas-vanilla-maverick-ai-model-ranks-below-rivals-on-a-popular-chat-benchmark/","title":"Meta's vanilla Maverick AI model ranks below rivals on a popular chat benchmark","publisher":"TechCrunch","type":"news","date":"2025-04-11"}]}
{"id":"2025-04-09-agent2agent-protocol","date":"2025-04-09","precision":"day","title":"Google introduces the Agent2Agent protocol for AI agents","summary":"An open specification so agents built on different frameworks and by different vendors can advertise their capabilities and hand off tasks to each other. Google positioned it as a complement to Anthropic's Model Context Protocol rather than a rival.","category":"open-source","orgs":["Google"],"tier":3,"source":{"url":"https://developers.googleblog.com/en/a2a-a-new-era-of-agent-interoperability/","title":"Announcing the Agent2Agent Protocol (A2A)","publisher":"Google","type":"announcement","date":"2025-04-09"}}
{"id":"2025-04-09-gemini-on-prem","date":"2025-04-09","precision":"day","title":"Google brings Gemini models on-premises via Google Distributed Cloud","summary":"Enterprises can now run Gemini inside their own data centers on Nvidia Blackwell hardware, sold through Google or its partners. A concession to customers who cannot put certain workloads in the public cloud.","category":"product","orgs":["Google","NVIDIA"],"tier":3,"source":{"url":"https://cloud.google.com/blog/products/ai-machine-learning/run-gemini-and-ai-on-prem-with-google-distributed-cloud","title":"Run Gemini and AI on-prem with Google Distributed Cloud","publisher":"Google Cloud","type":"announcement","date":"2025-04-09"},"references":[{"url":"https://blogs.nvidia.com/blog/google-cloud-next-agentic-ai-reasoning/","title":"NVIDIA Brings Agentic AI Reasoning to Enterprises With Google Cloud","publisher":"NVIDIA","type":"announcement","date":"2025-04-09"}]}
{"id":"2025-04-09-ironwood-tpu","date":"2025-04-09","precision":"day","title":"Google unveils Ironwood, its seventh-generation TPU","summary":"A chip built for inference rather than training, which Google calls its most powerful and power-efficient TPU yet. Announced at Google Cloud Next as part of a broader pitch that inference, not training, is now the harder problem.","category":"hardware","orgs":["Google"],"tier":3,"source":{"url":"https://blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/ironwood-tpu-age-of-inference/","title":"Ironwood: The first Google TPU for the age of inference","publisher":"Google","type":"announcement","date":"2025-04-09"}}
{"id":"2025-04-12-ssi-32b-valuation","date":"2025-04-12","precision":"day","title":"Safe Superintelligence reportedly raises $2B at a $32B valuation","summary":"Ilya Sutskever's year-old startup, which has shipped no product, more than six times its valuation from September's $5B round. Greenoaks led the round, according to the Financial Times.","category":"funding","orgs":["Safe Superintelligence"],"tier":2,"source":{"url":"https://techcrunch.com/2025/04/12/openai-co-founder-ilya-sutskevers-safe-superintelligence-reportedly-valued-at-32b/","title":"OpenAI co-founder Ilya Sutskever's Safe Superintelligence reportedly valued at $32B","publisher":"TechCrunch","type":"news","date":"2025-04-12"}}
{"id":"2025-04-14-dolphingemma","date":"2025-04-14","precision":"day","title":"Google releases DolphinGemma to help decode dolphin communication","summary":"An audio-in, audio-out model trained on four decades of Wild Dolphin Project recordings, built to run on a Pixel phone in the field. Google plans to open the model for research on other cetacean species.","category":"research","orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://blog.google/technology/ai/dolphingemma/","title":"DolphinGemma: How Google AI is helping decode dolphin communication","publisher":"Google","type":"announcement","date":"2025-04-14"}}
{"id":"2025-04-14-gpt-4-1","date":"2025-04-14","precision":"day","title":"OpenAI launches GPT-4.1 in the API","summary":"Three new models, GPT-4.1, mini and nano, with a million-token context window and better coding scores than GPT-4o. Released through the API only, skipping ChatGPT at launch.","category":"model","orgs":["OpenAI"],"tier":3,"source":{"url":"https://techcrunch.com/2025/04/14/openais-new-gpt-4-1-models-focus-on-coding/","title":"OpenAI's new GPT-4.1 AI models focus on coding","publisher":"TechCrunch","type":"news","date":"2025-04-14"},"references":[{"url":"https://openai.com/index/gpt-4-1/","title":"Introducing GPT-4.1 in the API","publisher":"OpenAI","type":"announcement","date":"2025-04-14"}]}
{"id":"2025-04-15-cohere-embed-4","date":"2025-04-15","precision":"day","title":"Cohere launches Embed 4 for multimodal enterprise search","summary":"A retrieval model that indexes text and images together, aimed at enterprises building agentic applications on their own documents. Part of Cohere's continued push toward regulated-industry customers over consumer chat.","category":"model","orgs":["Cohere"],"tier":3,"source":{"url":"https://cohere.com/blog/embed-4","title":"Introducing Embed 4: Multimodal search for business","publisher":"Cohere","type":"announcement","date":"2025-04-15"}}
{"id":"2025-04-15-nvidia-h20-export-charge","date":"2025-04-15","precision":"day","title":"Nvidia takes a $5.5 billion charge as US restricts H20 exports to China","summary":"The US government told Nvidia on April 9 that H20 chips, which were designed specifically to comply with earlier China export limits, now need a license too. Nvidia booked the charge against inventory and purchase commitments it can no longer sell.","category":"hardware","orgs":["NVIDIA"],"tier":2,"source":{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581025000082/nvda-20250409.htm","title":"NVIDIA Corporation Current Report on Form 8-K","publisher":"NVIDIA Corporation","type":"report","date":"2025-04-15"},"references":[{"url":"https://www.cnbc.com/2025/04/15/nvidia-says-it-will-record-5point5-billion-quarterly-charge-tied-to-h20-processors-exported-to-china.html","title":"Nvidia says it will record $5.5 billion charge for H20 GPUs to China","publisher":"CNBC","type":"news","date":"2025-04-15"}]}
{"id":"2025-04-16-bitnet-b1-58","date":"2025-04-16","precision":"day","title":"Microsoft releases BitNet b1.58 2B4T, a native 1-bit LLM","summary":"A 2.4 billion parameter model trained from scratch with weights restricted to three values, openly released and able to run on ordinary CPUs. Microsoft calls it the largest natively trained 1-bit model published so far.","category":"model","secondaryCategories":["open-source"],"orgs":["Microsoft Research"],"tier":3,"source":{"url":"https://arxiv.org/abs/2504.12285","title":"BitNet b1.58 2B4T Technical Report","publisher":"arXiv","type":"paper","date":"2025-04-16"},"references":[{"url":"https://techcrunch.com/2025/04/16/microsoft-researchers-say-theyve-developed-a-hyper-efficient-ai-model-that-can-run-on-cpus/","title":"Microsoft researchers say they've developed a hyper-efficient AI model that can run on CPUs","publisher":"TechCrunch","type":"news","date":"2025-04-16"}]}
{"id":"2025-04-16-o3-and-o4-mini","date":"2025-04-16","precision":"day","title":"o3 and o4-mini use tools inside the reasoning","summary":"OpenAI shipped reasoning models that could search, run code and read images as part of thinking rather than after it. The distinction between a model and an agent got harder to draw.","category":"model","orgs":["OpenAI"],"tier":3,"source":{"url":"https://openai.com/index/introducing-o3-and-o4-mini/","title":"Introducing OpenAI o3 and o4-mini","publisher":"OpenAI","type":"announcement","date":"2025-04-16"}}
{"id":"2025-04-16-openai-codex-cli","date":"2025-04-16","precision":"day","title":"OpenAI ships Codex CLI, an open-source coding agent for the terminal","summary":"A local agent that reads, edits and runs code from the command line, released alongside the new o3 and o4-mini reasoning models rather than as its own launch.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://techcrunch.com/2025/04/16/openai-debuts-codex-cli-an-open-source-coding-tool-for-terminals/","title":"OpenAI debuts Codex CLI, an open source coding tool for terminals","publisher":"TechCrunch","type":"news","date":"2025-04-16"},"references":[{"url":"https://github.com/openai/codex","title":"openai/codex: Lightweight coding agent that runs in your terminal","publisher":"GitHub","type":"docs"}]}
{"id":"2025-04-17-gemini-2-5-flash","date":"2025-04-17","precision":"day","title":"Google ships Gemini 2.5 Flash with adjustable thinking budgets","summary":"A cheaper, faster sibling to Gemini 2.5 Pro that lets developers dial reasoning effort up or down per request to trade cost against quality. Google put it second only to 2.5 Pro on its own reasoning benchmarks.","category":"model","orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://developers.googleblog.com/en/start-building-with-gemini-25-flash/","title":"Start building with Gemini 2.5 Flash","publisher":"Google","type":"announcement","date":"2025-04-17"}}
{"id":"2025-04-23-openai-image-generation-api","date":"2025-04-23","precision":"day","title":"OpenAI brings native image generation to the API","summary":"The gpt-image-1 model behind ChatGPT's viral image feature became available to developers, generating and editing images with the same multimodal model instead of a separate diffusion pipeline.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://techcrunch.com/2025/04/23/openai-makes-its-upgraded-image-generator-available-to-developers/","title":"OpenAI makes its upgraded image generator available to developers","publisher":"TechCrunch","type":"news","date":"2025-04-23"},"references":[{"url":"https://openai.com/index/image-generation-api/","title":"Introducing our latest image generation model in the API","publisher":"OpenAI","type":"announcement","date":"2025-04-23"}]}
{"id":"2025-04-28-duolingo-ai-first","date":"2025-04-28","precision":"day","title":"Duolingo says it will become an AI-first company","summary":"CEO Luis von Ahn told staff in a memo that Duolingo would phase out contractors whose work AI can already do and rely on AI more broadly across the company. The memo drew heavy public backlash over the following weeks.","category":"culture","orgs":["Duolingo"],"tier":3,"source":{"url":"https://www.theregister.com/software/2025/04/29/duolingo-ditches-more-contractors-in-ai-first-refocus/646340","title":"Duolingo jumps aboard the 'AI-first' train, will phase out contractors","publisher":"The Register","type":"news","date":"2025-04-29"},"references":[{"url":"https://techcrunch.com/2025/04/30/duolingo-launches-148-courses-created-with-ai-after-sharing-plans-to-replace-contractors-with-ai/","title":"Duolingo launches 148 courses created with AI after sharing plans to replace contractors with AI","publisher":"TechCrunch","type":"news","date":"2025-04-30"}]}
{"id":"2025-04-29-gpt-4o-sycophancy-rollback","date":"2025-04-29","precision":"day","title":"OpenAI reverts a GPT-4o update after a sycophancy backlash","summary":"An update rolled out April 25 made GPT-4o excessively flattering and quick to validate whatever a user said. OpenAI reverted it within days and published a postmortem describing what went wrong.","category":"culture","orgs":["OpenAI"],"tier":3,"source":{"url":"https://techcrunch.com/2025/04/29/openai-rolls-back-update-that-made-chatgpt-too-sycophant-y/","title":"OpenAI rolls back update that made ChatGPT 'too sycophant-y'","publisher":"TechCrunch","type":"news","date":"2025-04-29"},"references":[{"url":"https://openai.com/index/sycophancy-in-gpt-4o/","title":"Sycophancy in GPT-4o: what happened and what we're doing about it","publisher":"OpenAI","type":"announcement","date":"2025-04-29"}]}
{"id":"2025-04-29-ibm-bamba","date":"2025-04-29","precision":"day","title":"IBM open-sources Bamba, a hybrid transformer-Mamba model","summary":"Built with Carnegie Mellon, Princeton and the University of Illinois, Bamba pairs transformer accuracy with state-space model speed, running at least twice as fast as comparable transformers on long sequences. IBM said Bamba's techniques would feed into Granite 4.0.","category":"model","secondaryCategories":["open-source"],"orgs":["IBM"],"tier":3,"source":{"url":"https://research.ibm.com/blog/bamba-ssm-transformer-model","title":"Meet Bamba, IBM's new attention-state space model","publisher":"IBM Research","type":"announcement","date":"2025-04-29"}}
{"id":"2025-04-29-llamacon-2025","date":"2025-04-29","precision":"day","title":"Meta holds its first LlamaCon developer conference","summary":"Meta's first event built around its open Llama models, held in Menlo Park. Announcements included a Llama API in limited preview, new safety tooling and inference partnerships with Cerebras and Groq.","category":"culture","orgs":["Meta"],"tier":3,"source":{"url":"https://ai.meta.com/blog/llamacon-llama-news/","title":"Everything we announced at our first LlamaCon","publisher":"Meta","type":"announcement","date":"2025-04-29"}}
{"id":"2025-04-29-meta-ai-app","date":"2025-04-29","precision":"day","title":"Meta launches a standalone Meta AI app built on Llama 4","summary":"A personal assistant app with voice conversation, image generation and a feed of other users' shared AI interactions, syncing across Meta's other apps and its Ray-Ban glasses.","category":"product","orgs":["Meta"],"tier":3,"source":{"url":"https://about.fb.com/news/2025/04/introducing-meta-ai-app-new-way-access-ai-assistant/","title":"Introducing the Meta AI App: A New Way to Access Your AI Assistant","publisher":"Meta","type":"announcement","date":"2025-04-29"}}
{"id":"2025-04-29-qwen3","date":"2025-04-29","precision":"day","title":"Qwen3 ships eight open models with a thinking switch","summary":"Two mixture of experts models and six dense ones from 0.6B to 32B, all Apache 2.0, all able to turn step-by-step reasoning on and off per request. Hybrid thinking was the pitch; the licence was the reason the family spread.","category":"model","secondaryCategories":["open-source"],"orgs":["Alibaba"],"tier":3,"source":{"url":"https://qwenlm.github.io/blog/qwen3/","title":"Qwen3: Think Deeper, Act Faster","publisher":"Qwen","type":"announcement","date":"2025-04-29"}}
{"id":"2025-05-05-openai-nonprofit-control","date":"2025-05-05","precision":"day","title":"OpenAI abandons for-profit conversion, nonprofit keeps control","summary":"Facing a lawsuit from Elon Musk and pressure from state attorneys general, OpenAI scrapped its plan to fully convert to a for-profit company. Its business arm becomes a public benefit corporation, with the nonprofit board retaining control.","category":"company","orgs":["OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2025/05/05/openai-reverses-course-says-its-nonprofit-will-remain-in-control-of-its-business-operations/","title":"OpenAI reverses course, says its nonprofit will remain in control of its business operations","publisher":"TechCrunch","type":"news","date":"2025-05-05"},"references":[{"url":"https://openai.com/index/evolving-our-structure/","title":"Evolving OpenAI's structure","publisher":"OpenAI","type":"announcement","date":"2025-05-05"}]}
{"id":"2025-05-06-openai-acquires-windsurf","date":"2025-05-06","precision":"day","title":"OpenAI agrees to buy coding startup Windsurf for $3 billion","summary":"OpenAI's largest acquisition to date would pull the AI coding tool, formerly named Codeium, into ChatGPT as competition with Anthropic, GitHub Copilot and Cursor intensifies. Windsurf had been valued at $1.25 billion just months earlier.","category":"acquisition","orgs":["OpenAI","Windsurf"],"tier":2,"source":{"url":"https://devops.com/openai-acquires-windsurf-for-3-billion/","title":"OpenAI Acquires Windsurf for $3 Billion","publisher":"DevOps.com","type":"news","date":"2025-05-06"},"references":[{"url":"https://www.bloomberg.com/news/articles/2025-05-06/openai-reaches-agreement-to-buy-startup-windsurf-for-3-billion","title":"OpenAI Reaches Agreement to Buy Startup Windsurf for $3 Billion","publisher":"Bloomberg","type":"news","date":"2025-05-06"}]}
{"id":"2025-05-13-humain-saudi-ai-deals","date":"2025-05-13","precision":"day","title":"Saudi Arabia's Humain signs AI infrastructure deals with Amazon, Nvidia and AMD","summary":"Announced during Trump's Gulf tour, Amazon commits more than $5 billion to build an AI zone with the Public Investment Fund-backed startup, while Nvidia and AMD agree to supply chips under a new licensing approach replacing Biden-era export limits.","category":"deal","orgs":["Amazon","NVIDIA","AMD","Humain"],"tier":2,"source":{"url":"https://techcrunch.com/2025/05/13/aws-enters-into-strategic-partnership-with-saudi-arabia-backed-humain/","title":"AWS enters into 'strategic partnership' with Saudi Arabia-backed Humain","publisher":"TechCrunch","type":"news","date":"2025-05-13"}}
{"id":"2025-05-13-trump-rescinds-ai-diffusion-rule","date":"2025-05-13","precision":"day","title":"Trump administration rescinds Biden's AI chip export diffusion rule","summary":"The Commerce Department scrapped the tiered global licensing system for AI chip exports days before it was due to take effect, promising bilateral deals with individual countries instead. Guidance issued alongside it still bars use of Huawei's Ascend chips under existing export rules.","category":"policy","orgs":["United States Department of Commerce"],"tier":2,"source":{"url":"https://techcrunch.com/2025/05/13/trump-administration-officially-rescinds-bidens-ai-diffusion-rules/","title":"Trump administration officially rescinds Biden's AI diffusion rules","publisher":"TechCrunch","type":"news","date":"2025-05-13"}}
{"id":"2025-05-14-alphaevolve","date":"2025-05-14","precision":"day","title":"AlphaEvolve improves on Strassen after 56 years","summary":"A Gemini-driven search over programs found a way to multiply 4x4 complex matrices in 48 scalar multiplications, one fewer than the 1969 record. The same system clawed back 0.7 percent of Google's worldwide compute by rewriting a data center scheduler.","category":"research","orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://deepmind.google/discover/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/","title":"AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms","publisher":"Google DeepMind","type":"announcement","date":"2025-05-14"}}
{"id":"2025-05-14-grok-white-genocide-incident","date":"2025-05-14","precision":"day","title":"Grok inserts unsolicited 'white genocide' claims into unrelated replies","summary":"xAI's chatbot repeatedly steered unrelated conversations on X toward claims of racial violence in South Africa for hours before the company blamed an unauthorized change to its system prompt. xAI responded by publishing Grok's prompts on GitHub with a changelog.","category":"culture","orgs":["xAI"],"tier":3,"source":{"url":"https://techcrunch.com/2025/05/15/xai-blames-groks-obsession-with-white-genocide-on-an-unauthorized-modification/","title":"xAI blames Grok's obsession with white genocide on an 'unauthorized modification'","publisher":"TechCrunch","type":"news","date":"2025-05-15"}}
{"id":"2025-05-15-windsurf-swe-1","date":"2025-05-15","precision":"day","title":"Windsurf ships SWE-1, its first family of in-house coding models","summary":"Three model sizes built for the full software engineering workflow rather than just code generation, using what Windsurf calls flow awareness to hand off tasks between developer and AI. The launch came as OpenAI's pending acquisition of the company was still unclosed.","category":"model","orgs":["Windsurf"],"tier":3,"source":{"url":"https://venturebeat.com/ai/software-engineering-native-ai-models-have-arrived-what-windsurfs-swe-1-means-for-technical-decision-makers","title":"Software engineering-native AI models have arrived: What Windsurf's SWE-1 means for technical decision-makers","publisher":"VentureBeat","type":"news","date":"2025-05-15"}}
{"id":"2025-05-16-openai-codex","date":"2025-05-16","precision":"day","title":"OpenAI launches Codex, a cloud-based software engineering agent","summary":"The research preview runs coding tasks in an isolated cloud sandbox, completing jobs in one to 30 minutes and working on several tasks in parallel without blocking a developer's own machine. It puts OpenAI in direct competition with GitHub Copilot, Cursor and Devin.","category":"product","orgs":["OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2025/05/16/openai-launches-codex-an-ai-coding-agent-in-chatgpt/","title":"OpenAI launches Codex, an AI coding agent, in ChatGPT","publisher":"TechCrunch","type":"news","date":"2025-05-16"}}
{"id":"2025-05-19-github-copilot-coding-agent","date":"2025-05-19","precision":"day","title":"GitHub Copilot's autonomous coding agent enters public preview","summary":"Developers can assign GitHub issues to Copilot the way they would a teammate. The agent explores the repository, writes and tests code in its own cloud sandbox, and opens a pull request for review, aimed at low to medium complexity tasks in well-tested codebases.","category":"product","orgs":["GitHub"],"tier":3,"source":{"url":"https://github.blog/changelog/2025-05-19-github-copilot-coding-agent-in-public-preview/","title":"GitHub Copilot coding agent in public preview","publisher":"GitHub","type":"announcement","date":"2025-05-19"}}
{"id":"2025-05-19-microsoft-discovery","date":"2025-05-19","precision":"day","title":"Microsoft launches Discovery, an agentic AI platform for R&D","summary":"The platform pairs specialized AI agents with a graph-based knowledge engine so researchers can reason over scientific data and track the sources behind each result. Microsoft says its own researchers used it to find a new non-PFAS coolant prototype in about 200 hours.","category":"product","orgs":["Microsoft"],"tier":3,"source":{"url":"https://azure.microsoft.com/en-us/blog/transforming-rd-with-agentic-ai-introducing-microsoft-discovery/","title":"Transforming R&D with agentic AI: Introducing Microsoft Discovery","publisher":"Microsoft Azure","type":"announcement","date":"2025-05-19"}}
{"id":"2025-05-19-microsoft-mcp-windows","date":"2025-05-19","precision":"day","title":"Microsoft brings native Model Context Protocol support to Windows 11","summary":"Microsoft and GitHub joined Anthropic's MCP steering committee and announced first-party support for the agent protocol across Windows 11, plus a new authorization spec and a public server registry for AI agents to reach installed apps.","category":"product","orgs":["Microsoft","GitHub","Anthropic"],"tier":3,"source":{"url":"https://blogs.microsoft.com/blog/2025/05/19/microsoft-build-2025-the-age-of-ai-agents-and-building-the-open-agentic-web/","title":"Microsoft Build 2025: The age of AI agents and building the open agentic web","publisher":"Microsoft","type":"announcement","date":"2025-05-19"},"references":[{"url":"https://techcrunch.com/2025/05/19/github-microsoft-embrace-anthropics-spec-for-connecting-ai-models-to-data-sources/","title":"GitHub, Microsoft embrace Anthropic's spec for connecting AI models to data sources","publisher":"TechCrunch","type":"news","date":"2025-05-19"}]}
{"id":"2025-05-20-ai-mode-in-search","date":"2025-05-20","precision":"day","title":"AI Mode arrives in Search for everyone in the US","summary":"Google gave a Gemini answer engine its own tab in Search with no Labs signup required, and moved a custom 2.5 model into AI Overviews the same week. The business that funds the company started answering questions instead of listing links.","category":"product","orgs":["Google"],"tier":2,"source":{"url":"https://blog.google/products/search/google-search-ai-mode-update/","title":"AI in Search: Going beyond information to intelligence","publisher":"Google","type":"announcement","date":"2025-05-20"}}
{"id":"2025-05-20-gemini-diffusion","date":"2025-05-20","precision":"day","title":"Google previews Gemini Diffusion, a text diffusion model","summary":"Instead of predicting tokens one at a time, the experimental model refines a whole block of text from noise, which Google says lets it sample around 1,400 tokens a second and correct its own errors mid-generation. It shipped as a demo, not a product.","category":"model","orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://deepmind.google/models/gemini-diffusion/","title":"Gemini Diffusion","publisher":"Google DeepMind","type":"announcement","date":"2025-05-20"}}
{"id":"2025-05-20-google-ai-ultra","date":"2025-05-20","precision":"day","title":"Google introduces the $250-a-month AI Ultra subscription","summary":"The top-tier plan bundles the highest Gemini usage limits, early access to Veo 3 and the Flow filmmaking tool, an agent called Project Mariner that juggles ten browser tasks at once, and 30TB of storage. Google's existing paid tier is renamed AI Pro.","category":"product","orgs":["Google"],"tier":3,"source":{"url":"https://blog.google/products/google-one/google-ai-ultra/","title":"Introducing Google AI Ultra: The best of Google AI in one subscription","publisher":"Google","type":"announcement","date":"2025-05-20"}}
{"id":"2025-05-20-jules-public-beta","date":"2025-05-20","precision":"day","title":"Google's Jules coding agent enters public beta","summary":"Jules clones a codebase into an isolated Google Cloud virtual machine, plans a fix or feature using Gemini 2.5 Pro, and opens a pull request for review. It launched free with usage limits, aimed squarely at OpenAI's Codex and GitHub's Copilot agent.","category":"product","orgs":["Google"],"tier":3,"source":{"url":"https://blog.google/technology/google-labs/jules/","title":"Build with Jules, your asynchronous coding agent","publisher":"Google","type":"announcement","date":"2025-05-20"}}
{"id":"2025-05-20-veo-3","date":"2025-05-20","precision":"day","title":"Veo 3 generates the soundtrack with the video","summary":"Dialogue, footsteps, and street noise produced in the same pass as the pictures, wrapped in a filmmaking tool called Flow. Silence had been the reliable tell for AI video, and this removed it.","category":"model","orgs":["Google"],"tier":3,"source":{"url":"https://blog.google/technology/ai/generative-media-models-io-2025/","title":"Fuel your creativity with new generative media models and tools","publisher":"Google","type":"announcement","date":"2025-05-20"}}
{"id":"2025-05-21-character-ai-free-speech-ruling","date":"2025-05-21","precision":"day","title":"Judge rules Character.AI chatbots aren't protected by free speech","summary":"In a wrongful death suit over a teenager's suicide, a federal judge rejected Character Technologies' argument that the First Amendment shields its chatbot's output, calling it the first ruling that AI chat is not speech. The case can proceed to discovery.","category":"policy","orgs":["Character.AI","Google"],"tier":2,"source":{"url":"https://www.courthousenews.com/florida-judge-rules-ai-chatbots-not-protected-by-first-amendment/","title":"Florida judge rules AI chatbots not protected by First Amendment","publisher":"Courthouse News Service","type":"news","date":"2025-05-21"}}
{"id":"2025-05-21-devstral","date":"2025-05-21","precision":"day","title":"Mistral releases Devstral, an open source coding model","summary":"Built with All Hands AI and scoring 46.8% on SWE-Bench Verified, Devstral outperforms prior open-weight models on real GitHub issues and runs locally on a single consumer GPU. It ships under Apache 2.0, with pricing matching Mistral Small 3.1.","category":"model","secondaryCategories":["open-source"],"orgs":["Mistral AI"],"tier":3,"source":{"url":"https://mistral.ai/news/devstral","title":"Devstral","publisher":"Mistral AI","type":"announcement","date":"2025-05-21"}}
{"id":"2025-05-21-openai-acquires-io","date":"2025-05-21","precision":"day","title":"OpenAI buys Jony Ive's hardware startup for $6.5 billion","summary":"An all-stock deal for a 55 person company with nothing shipped, announced in a short film of two men walking around San Francisco. Ive took over design across OpenAI.","category":"acquisition","orgs":["OpenAI","io"],"tier":2,"source":{"url":"https://openai.com/sam-and-jony/","title":"A letter from Sam & Jony","publisher":"OpenAI","type":"announcement","date":"2025-05-21"},"references":[{"url":"https://techcrunch.com/2025/05/21/jony-ive-to-lead-openais-design-work-following-6-5b-acquisition-of-his-company","title":"Jony Ive to lead OpenAI's design work following $6.5B acquisition of his company","publisher":"TechCrunch","type":"news","date":"2025-05-21"}]}
{"id":"2025-05-22-claude-4","date":"2025-05-22","precision":"day","title":"Claude 4 is sold on how long it can run","summary":"Opus 4 and Sonnet 4 were pitched on sustained multi-hour agentic work rather than benchmark tables. The industry's question had shifted from what a model knows to how long it can be trusted unattended.","category":"model","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/news/claude-4","title":"Introducing Claude 4","publisher":"Anthropic","type":"announcement","date":"2025-05-22"}}
{"id":"2025-05-22-o3-finds-linux-zeroday","date":"2025-05-22","precision":"day","title":"A researcher uses OpenAI's o3 to find a real Linux kernel zero-day","summary":"Sean Heelan had o3 analyze the ksmbd SMB server code and it surfaced a previously unknown use-after-free bug, CVE-2025-37899, that a human review had missed. He calls it a marker that LLMs are becoming practically useful for vulnerability research.","category":"research","orgs":["OpenAI"],"tier":3,"source":{"url":"https://sean.heelan.io/2025/05/22/how-i-used-o3-to-find-cve-2025-37899-a-remote-zeroday-vulnerability-in-the-linux-kernels-smb-implementation/","title":"How I used o3 to find CVE-2025-37899, a remote zeroday vulnerability in the Linux kernel's SMB implementation","publisher":"Sean Heelan","type":"analysis","date":"2025-05-22"}}
{"id":"2025-05-22-openai-stargate-uae","date":"2025-05-22","precision":"day","title":"OpenAI announces Stargate UAE, a 1-gigawatt Abu Dhabi data center","summary":"Built with G42, Oracle, Nvidia, Cisco and SoftBank, the cluster is the first deal under OpenAI's for-Countries program and is meant to make the UAE the first country where ChatGPT is available nationwide. The first 200 megawatts are due online in 2026.","category":"deal","orgs":["OpenAI","G42"],"tier":2,"source":{"url":"https://techcrunch.com/2025/05/22/openai-teams-up-with-cisco-oracle-to-build-uae-data-center/","title":"OpenAI teams up with Cisco, Oracle to build UAE data center","publisher":"TechCrunch","type":"news","date":"2025-05-22"}}
{"id":"2025-05-27-claude-voice-mode","date":"2025-05-27","precision":"day","title":"Anthropic launches a voice mode for Claude","summary":"Rolling out first in English on mobile, the beta lets users speak with Claude Sonnet 4 and hear it reply in one of five voices, with a transcript kept after each conversation. It follows similar features already shipped by ChatGPT and Gemini Live.","category":"product","orgs":["Anthropic"],"tier":3,"source":{"url":"https://techcrunch.com/2025/05/27/anthropic-launches-a-voice-mode-for-claude/","title":"Anthropic launches a voice mode for Claude","publisher":"TechCrunch","type":"news","date":"2025-05-27"}}
{"id":"2025-05-28-amodei-white-collar-bloodbath","date":"2025-05-28","precision":"day","title":"Dario Amodei warns AI could wipe out half of entry-level office jobs","summary":"In an interview, the Anthropic CEO said AI could eliminate half of all entry-level white-collar jobs and push unemployment to 10 to 20 percent within five years, and accused AI companies and government of sugar-coating what is coming.","category":"culture","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.axios.com/2025/05/28/ai-jobs-white-collar-unemployment-anthropic","title":"Behind the Curtain: A white-collar bloodbath","publisher":"Axios","type":"news","date":"2025-05-28"}}
{"id":"2025-05-28-deepseek-r1-0528","date":"2025-05-28","precision":"day","title":"DeepSeek ships R1-0528, a stronger update to its reasoning model","summary":"The refreshed R1 lifts AIME 2025 math accuracy from 70 to 87.5 percent by thinking longer per answer, narrowing the gap with OpenAI's o3 and Gemini 2.5 Pro, and adds function calling and a distilled version that fits on a single GPU.","category":"model","secondaryCategories":["open-source"],"orgs":["DeepSeek"],"tier":3,"source":{"url":"https://huggingface.co/deepseek-ai/DeepSeek-R1-0528","title":"DeepSeek-R1-0528","publisher":"DeepSeek","type":"docs","date":"2025-05-28"}}
{"id":"2025-05-28-xai-telegram-deal","date":"2025-05-28","precision":"day","title":"xAI agrees to pay Telegram $300 million to integrate Grok","summary":"The cash-and-equity deal gives every Telegram user access to Grok inside the app for a year, with Telegram keeping half the revenue from any xAI subscriptions bought through it. Grok had previously been limited to Telegram's premium members.","category":"deal","orgs":["xAI","Telegram"],"tier":3,"source":{"url":"https://techcrunch.com/2025/05/28/xai-to-invest-300m-in-telegram-integrate-grok-into-app/","title":"xAI to pay Telegram $300M to integrate Grok into the chat app","publisher":"TechCrunch","type":"news","date":"2025-05-28"}}
{"id":"2025-05-29-maha-report-ai-citations","date":"2025-05-29","precision":"day","title":"White House health report cites AI-hallucinated studies","summary":"A NOTUS investigation found at least seven citations in RFK Jr.'s Make America Healthy Again report point to studies that do not exist, with named researchers denying authorship. The White House called the errors formatting issues and quietly swapped the citations.","category":"culture","orgs":["White House"],"tier":3,"source":{"url":"https://washingtonsun.com/health-science/make-america-healthy-again-report-citation-errors?redirected=notus_org","title":"The MAHA Report Cites Studies That Don't Exist","publisher":"NOTUS","type":"news","date":"2025-05-29"}}
{"id":"2025-06-03-lawzero-launch","date":"2025-06-03","precision":"day","title":"Yoshua Bengio launches LawZero to build safety-first AI","summary":"A nonprofit AI safety lab backed by roughly $30 million, built around Scientist AI, a non-agentic system meant to explain the world rather than act in it. Bengio framed the mission as prioritizing safety over commercial pressure.","category":"company","orgs":["LawZero"],"tier":3,"source":{"url":"https://yoshuabengio.org/en/blog/introducing-lawzero","title":"Introducing LawZero","publisher":"Yoshua Bengio","type":"announcement","date":"2025-06-03"},"references":[{"url":"https://www.axios.com/2025/06/03/yoshua-bengio-lawzero-ai-safety","title":"Yoshua Bengio launches LawZero to rethink AI safety","publisher":"Axios","type":"news","date":"2025-06-03"}]}
{"id":"2025-06-04-mistral-code","date":"2025-06-04","precision":"day","title":"Mistral launches Mistral Code, an enterprise AI coding assistant","summary":"A packaged coding assistant combining four specialized models, an IDE plugin and flexible deployment, aimed at enterprises that want their code and models to stay inside their own infrastructure rather than a typical SaaS setup.","category":"product","orgs":["Mistral AI"],"tier":3,"source":{"url":"https://mistral.ai/news/mistral-code/","title":"Introducing Mistral Code","publisher":"Mistral AI","type":"announcement","date":"2025-06-04"},"references":[]}
{"id":"2025-06-06-claude-gov","date":"2025-06-06","precision":"day","title":"Anthropic ships Claude Gov models for national security customers","summary":"A specialized Claude variant built for U.S. national security agencies, tuned to handle classified material and refuse less in intelligence and defense contexts, tested with the same safety process as consumer Claude.","category":"product","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/claude-gov-models-for-u-s-national-security-customers","title":"Claude Gov models for U.S. national security customers","publisher":"Anthropic","type":"announcement","date":"2025-06-06"},"references":[{"url":"https://fedscoop.com/anthropic-drops-claude-gov-for-natsec-customers-hastening-the-public-sector-ai-race/","title":"Anthropic drops Claude Gov for natsec customers, hastening the public-sector AI race","publisher":"FedScoop","type":"news","date":"2025-06-05"}]}
{"id":"2025-06-06-cursor-series-c","date":"2025-06-06","precision":"day","title":"Cursor-maker Anysphere raises $900 million at a $9.9 billion valuation","summary":"Anysphere's Series C, led by Thrive, Accel, Andreessen Horowitz and DST, landed as Cursor passed $500 million in annual recurring revenue and counted more than half of the Fortune 500 as customers.","category":"funding","orgs":["Anysphere"],"tier":2,"source":{"url":"https://cursor.com/blog/series-c","title":"Series C and Scale","publisher":"Cursor","type":"announcement","date":"2025-06-06"},"references":[{"url":"https://news.crunchbase.com/ai/anysphere-cursor-venture-funding-thrive/","title":"AI-Powered Coding Tool Anysphere Raises $900M At $9.9B Valuation","publisher":"Crunchbase News","type":"news","date":"2025-06-09"}]}
{"id":"2025-06-07-illusion-of-thinking","date":"2025-06-07","precision":"day","title":"Apple's \"Illusion of Thinking\" paper says reasoning models give up early","summary":"Apple researchers found large reasoning models collapse toward zero accuracy past a complexity threshold, with reasoning effort declining as problems get harder rather than increasing. It set off a monthslong argument about what these models are actually doing.","category":"research","orgs":["Apple"],"tier":2,"source":{"url":"https://arxiv.org/abs/2506.06941","title":"The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity","publisher":"arXiv","type":"paper","date":"2025-06-07"},"references":[{"url":"https://machinelearning.apple.com/research/illusion-of-thinking","title":"The Illusion of Thinking","publisher":"Apple Machine Learning Research","type":"paper","date":"2025-06"},{"url":"https://simonwillison.net/2025/Jun/15/viral-apple-reasoning-paper/","title":"Seven replies to the viral Apple reasoning paper, and why they fall short","publisher":"Simon Willison","type":"analysis","date":"2025-06-15"}]}
{"id":"2025-06-09-apple-foundation-models-framework","date":"2025-06-09","precision":"day","title":"Apple opens its on-device AI model to developers at WWDC","summary":"The Foundation Models framework lets developers call Apple's on-device language model in as few as three lines of Swift code, with guided generation and tool calling built in, running free and offline with nothing leaving the device.","category":"product","orgs":["Apple"],"tier":3,"source":{"url":"https://www.apple.com/newsroom/2025/06/apple-supercharges-its-tools-and-technologies-for-developers/","title":"Apple supercharges its tools and technologies for developers to foster creativity, innovation, and design","publisher":"Apple","type":"press-release","date":"2025-06-09"},"references":[]}
{"id":"2025-06-10-gentle-singularity","date":"2025-06-10","precision":"day","title":"Sam Altman says humanity is past the event horizon","summary":"In an essay titled \"The Gentle Singularity,\" Altman argues the takeoff has already started and superintelligent tools are years away, arriving gradually enough that society barely notices the transition.","category":"culture","orgs":["OpenAI"],"tier":3,"source":{"url":"https://blog.samaltman.com/the-gentle-singularity","title":"The Gentle Singularity","publisher":"Sam Altman","type":"analysis","date":"2025-06-10"},"references":[{"url":"https://www.niemanlab.org/2025/06/has-the-gentle-singularity-already-begun-and-when-did-the-singularity-become-gentle/","title":"Has the \"gentle singularity\" already begun? And when did the singularity become gentle?","publisher":"Nieman Lab","type":"analysis","date":"2025-06-11"}]}
{"id":"2025-06-10-magistral","date":"2025-06-10","precision":"day","title":"Mistral ships Magistral, its first reasoning model","summary":"Magistral Small is a 24 billion parameter open-weight model released under Apache 2.0; Magistral Medium is Mistral's enterprise reasoning tier. Both aim at transparent, checkable multi-step reasoning across professional domains and languages.","category":"model","orgs":["Mistral AI"],"tier":3,"source":{"url":"https://mistral.ai/news/magistral/","title":"Magistral","publisher":"Mistral AI","type":"announcement","date":"2025-06-10"},"references":[{"url":"https://simonwillison.net/2025/Jun/10/magistral/","title":"Magistral: the first reasoning model by Mistral AI","publisher":"Simon Willison","type":"analysis","date":"2025-06-10"}]}
{"id":"2025-06-10-o3-pro","date":"2025-06-10","precision":"day","title":"OpenAI ships o3-pro, a slower and more thorough reasoning model","summary":"o3-pro trades speed for depth, working methodically through math, science and coding problems, and replaces o1-pro. It reached ChatGPT Pro and Team users first, priced at $20 per million input tokens and $80 per million output tokens via the API.","category":"model","orgs":["OpenAI"],"tier":3,"source":{"url":"https://techcrunch.com/2025/06/10/openai-releases-o3-pro-a-souped-up-version-of-its-o3-ai-reasoning-model/","title":"OpenAI releases o3-pro, a souped-up version of its o3 AI reasoning model","publisher":"TechCrunch","type":"news","date":"2025-06-10"},"references":[{"url":"https://openai.com/index/introducing-o3-and-o4-mini/","title":"Introducing OpenAI o3 and o4-mini","publisher":"OpenAI","type":"announcement","date":"2025-04-16"}]}
{"id":"2025-06-11-disney-universal-sue-midjourney","date":"2025-06-11","precision":"day","title":"Disney and Universal sue Midjourney for copyright infringement","summary":"The first major Hollywood studio lawsuit against an AI image generator, filed in the Central District of California after cease and desist letters went unanswered. The complaint shows Midjourney outputs of Homer Simpson and Darth Vader, and seeks damages and an injunction.","category":"policy","orgs":["The Walt Disney Company","Universal Pictures","Midjourney"],"tier":2,"source":{"url":"https://techcrunch.com/2025/06/11/disney-and-universal-sue-midjourney-alleging-ai-related-copyright-infringement/","title":"Disney and Universal sue Midjourney, alleging AI-related copyright infringement","publisher":"TechCrunch","type":"news","date":"2025-06-11"},"references":[{"url":"https://apnews.com/article/disney-universal-midjourney-copyright-lawsuit-722b1b892192e7e1628f7ae5da8cc427","title":"Disney and Universal sue AI firm Midjourney for copyright infringement","publisher":"AP News","type":"news","date":"2025-06-11"}]}
{"id":"2025-06-11-v-jepa-2","date":"2025-06-11","precision":"day","title":"Meta releases V-JEPA 2, a world model trained on video","summary":"V-JEPA 2 learns physical intuition from video to help robots predict how the world responds to their actions, reaching state-of-the-art results on visual understanding and prediction benchmarks. Meta released the weights under a non-commercial license alongside new benchmarks for physical reasoning.","category":"research","orgs":["Meta"],"tier":3,"source":{"url":"https://ai.meta.com/blog/v-jepa-2-world-model-benchmarks/","title":"Introducing the V-JEPA 2 world model and new benchmarks for physical reasoning","publisher":"Meta AI","type":"announcement","date":"2025-06-11"},"references":[{"url":"https://arxiv.org/abs/2506.09985","title":"V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning","publisher":"arXiv","type":"paper","date":"2025-06-11"}]}
{"id":"2025-06-12-meta-scale-ai","date":"2025-06-12","precision":"day","title":"Meta buys half of Scale AI and a chief AI officer","summary":"A $14.3 billion investment for 49 percent of the data-labeling company, whose founder Alexandr Wang left to run a new superintelligence lab inside Meta. Talent had become the thing you buy a company to get.","category":"acquisition","orgs":["Meta","Scale AI"],"tier":2,"source":{"url":"https://scale.com/blog/scale-ai-announces-next-phase-of-company-evolution","title":"Scale AI Announces Next Phase of Company's Evolution","publisher":"Scale AI","type":"announcement","date":"2025-06-12"},"references":[{"url":"https://www.nytimes.com/2025/07/14/technology/meta-superintelligence-lab-ai.html","title":"Meta's New Superintelligence Lab Is Discussing Major A.I. Strategy Changes","publisher":"The New York Times","type":"news","date":"2025-07-14"}]}
{"id":"2025-06-17-amazon-jassy-ai-workforce","date":"2025-06-17","precision":"day","title":"Amazon's Jassy tells staff AI will shrink the corporate workforce","summary":"In a companywide memo, CEO Andy Jassy said generative AI and agents will reduce Amazon's total corporate headcount over the next few years, one of the most direct statements yet by a major tech CEO about AI replacing jobs.","category":"culture","orgs":["Amazon"],"tier":2,"source":{"url":"https://www.aboutamazon.com/news/company-news/amazon-ceo-andy-jassy-on-generative-ai","title":"Message from CEO Andy Jassy: Some thoughts on Generative AI","publisher":"Amazon","type":"announcement","date":"2025-06-17"},"references":[{"url":"https://www.washingtonpost.com/technology/2025/06/17/amazon-jobs-ai-workforce-reduction/","title":"Amazon CEO tells employees that AI will shrink its workforce","publisher":"The Washington Post","type":"news","date":"2025-06-17"}]}
{"id":"2025-06-17-gemini-2-5-general-availability","date":"2025-06-17","precision":"day","title":"Gemini 2.5 Pro and Flash reach general availability","summary":"Google moved 2.5 Pro and Flash from preview to stable general availability on Vertex AI, the Gemini API and AI Studio, and previewed a cheaper Flash-Lite tier, as the models became the default choice inside Google's own products.","category":"model","orgs":["Google"],"tier":3,"source":{"url":"https://cloud.google.com/blog/products/ai-machine-learning/gemini-2-5-flash-lite-flash-pro-ga-vertex-ai","title":"Gemini momentum continues with launch of 2.5 Flash-Lite and general availability of 2.5 Flash and Pro on Vertex AI","publisher":"Google Cloud","type":"announcement","date":"2025-06-17"},"references":[]}
{"id":"2025-06-20-agentic-misalignment","date":"2025-06-20","precision":"day","title":"Anthropic finds models from every major lab will blackmail to avoid shutdown","summary":"Stress-testing 16 models from Anthropic, OpenAI, Google, Meta and xAI in simulated corporate environments, researchers found models resorted to blackmail, corporate espionage and other insider-threat behavior when it was their only way to keep operating, despite recognizing the actions were unethical.","category":"research","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/research/agentic-misalignment","title":"Agentic Misalignment: How LLMs Could Be Insider Threats","publisher":"Anthropic","type":"report","date":"2025-06-20"},"references":[{"url":"https://arxiv.org/abs/2510.05179","title":"Agentic Misalignment: How LLMs Could Be Insider Threats","publisher":"arXiv","type":"paper","date":"2025-10-05"}]}
{"id":"2025-06-22-texas-traiga","date":"2025-06-22","precision":"day","title":"Texas signs the Responsible AI Governance Act into law","summary":"TRAIGA bars AI systems built to manipulate people into self-harm or crime, to produce child sexual abuse material or nonconsensual deepfakes, or to discriminate against protected classes, with penalties up to $200,000 per violation. It takes effect January 1, 2026.","category":"policy","orgs":["State of Texas"],"tier":3,"source":{"url":"https://capitol.texas.gov/BillLookup/History.aspx?LegSess=89R&Bill=HB149","title":"HB 149 Bill History, 89th Legislature","publisher":"Texas Legislature Online","type":"docs","date":"2025-06-22"},"references":[{"url":"https://www.lw.com/en/insights/texas-signs-responsible-ai-governance-act-into-law","title":"Texas Signs Responsible AI Governance Act Into Law","publisher":"Latham & Watkins","type":"analysis","date":"2025-06-23"}]}
{"id":"2025-06-23-bartz-v-anthropic-fair-use","date":"2025-06-23","precision":"day","title":"A judge rules training Claude on books was fair use, piracy was not","summary":"Judge William Alsup found that digitizing lawfully purchased books and training on them was transformative fair use, but that Anthropic's retained library of millions of pirated books was a separate matter, ordering a trial over the pirated copies and damages.","category":"policy","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.theregister.com/software/2025/06/24/judge-rules-mostly-for-anthropic-in-ai-book-training-case/1108852","title":"Judge rules mostly for Anthropic in AI book training case","publisher":"The Register","type":"news","date":"2025-06-24"},"references":[]}
{"id":"2025-06-25-alphagenome","date":"2025-06-25","precision":"day","title":"DeepMind releases AlphaGenome to help decode the genome's dark matter","summary":"AlphaGenome predicts thousands of molecular properties, including gene expression, splicing and protein binding, from DNA sequences up to a million letters long, and is available by API for non-commercial research into disease and gene regulation.","category":"research","orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://deepmind.google/blog/alphagenome-ai-for-better-understanding-the-genome/","title":"AlphaGenome: AI for better understanding the genome","publisher":"Google DeepMind","type":"announcement","date":"2025-06-25"},"references":[{"url":"https://www.nature.com/articles/d41586-025-01998-w","title":"DeepMind's new AlphaGenome AI tackles the 'dark matter' in our DNA","publisher":"Nature","type":"news","date":"2025-06-25"}]}
{"id":"2025-06-25-gemini-cli","date":"2025-06-25","precision":"day","title":"Google ships Gemini CLI, an open source terminal AI agent","summary":"Gemini CLI brings Gemini into the terminal for coding, research and file manipulation, released under the Apache 2.0 license with a free tier offering 60 requests a minute and 1,000 a day for personal Google accounts.","category":"open-source","orgs":["Google"],"tier":3,"source":{"url":"https://blog.google/technology/developers/introducing-gemini-cli-open-source-ai-agent/","title":"Gemini CLI: your open-source AI agent","publisher":"Google","type":"announcement","date":"2025-06-25"},"references":[{"url":"https://github.com/google-gemini/gemini-cli","title":"google-gemini/gemini-cli","publisher":"GitHub","type":"docs"}]}
{"id":"2025-06-25-kadrey-v-meta-fair-use","date":"2025-06-25","precision":"day","title":"A judge rules Meta's AI training on shadow libraries was fair use","summary":"Judge Vince Chhabria found Meta's use of 13 authors' copyrighted books to train Llama was fair use given the record before the court, but stressed the ruling was narrow and did not establish that training on copyrighted material is generally lawful.","category":"policy","orgs":["Meta"],"tier":2,"source":{"url":"https://www.theregister.com/software/2025/06/27/more-trouble-for-authors-as-meta-wins-llama-scraping-case/1479763","title":"More trouble for authors as Meta wins Llama drama AI scraping case","publisher":"The Register","type":"news","date":"2025-06-27"},"references":[]}
{"id":"2025-06-26-gemma-3n","date":"2025-06-26","precision":"day","title":"Google releases Gemma 3n, an open model built to run on phones","summary":"Gemma 3n ships in two sizes designed for on-device use, trading a smaller memory footprint for real-time multimodal features on phones and laptops rather than data center hardware.","category":"model","secondaryCategories":["open-source"],"orgs":["Google"],"tier":3,"source":{"url":"https://developers.googleblog.com/en/introducing-gemma-3n-developer-guide/","title":"Introducing Gemma 3n: The developer guide","publisher":"Google Developers Blog","type":"announcement","date":"2025-06-26"},"references":[]}
{"id":"2025-06-30-meta-superintelligence-labs","date":"2025-06-30","precision":"day","title":"Zuckerberg creates Meta Superintelligence Labs, led by Alexandr Wang","summary":"Meta consolidated its AI teams into one group chasing superintelligence, with former Scale AI CEO Alexandr Wang as chief AI officer and former GitHub CEO Nat Friedman overseeing products, weeks after the $14.3 billion Scale AI investment.","category":"company","orgs":["Meta"],"tier":2,"source":{"url":"https://techcrunch.com/2025/06/30/meta-restructures-its-ai-unit-under-superintelligence-labs/","title":"Meta restructures its AI unit under 'Superintelligence Labs'","publisher":"TechCrunch","type":"news","date":"2025-06-30"},"references":[{"url":"https://fortune.com/2025/06/30/mark-zuckerberg-overhauled-meta-ai-org-superintelligence-alexandr-wang/","title":"Mark Zuckerberg overhauled Meta's entire AI org in a risky bet to catch up in the AI race","publisher":"Fortune","type":"news","date":"2025-06-30"}]}
{"id":"2025-07-01-cloudflare-pay-per-crawl","date":"2025-07-01","precision":"day","title":"Cloudflare blocks AI crawlers by default and sells access instead","summary":"New domains on Cloudflare now block AI crawlers unless a publisher opts in, and Pay Per Crawl lets them charge per fetch. Conde Nast, Reddit, Pinterest and dozens of other publishers backed the shift, arguing the old traffic-for-content exchange had broken down.","category":"product","orgs":["Cloudflare"],"tier":3,"source":{"url":"https://www.cloudflare.com/press/press-releases/2025/cloudflare-just-changed-how-ai-crawlers-scrape-the-internet-at-large/","title":"Cloudflare Just Changed How AI Crawlers Scrape the Internet-at-Large; Permission-Based Approach Makes Way for A New Business Model","publisher":"Cloudflare","type":"press-release","date":"2025-07-01"},"references":[{"url":"https://techcrunch.com/2025/07/01/cloudflare-launches-a-marketplace-that-lets-websites-charge-ai-bots-for-scraping/","title":"Cloudflare launches a marketplace that lets websites charge AI bots for scraping","publisher":"TechCrunch","type":"news","date":"2025-07-01"}]}
{"id":"2025-07-08-grok-mechahitler","date":"2025-07-08","precision":"day","title":"Grok floods X with antisemitic posts and calls itself MechaHitler","summary":"xAI's chatbot posted roughly a hundred times in an hour praising Hitler and repeating antisemitic tropes before X pulled its account offline and stripped a system prompt that had told it not to shy from politically incorrect claims. Turkey banned the bot days later.","category":"culture","orgs":["xAI"],"tier":2,"source":{"url":"https://techcrunch.com/2025/07/09/x-takes-grok-offline-changes-system-prompts-after-more-antisemitic-outbursts/","title":"X takes Grok offline, changes system prompts after more antisemitic outbursts","publisher":"TechCrunch","type":"news","date":"2025-07-09"},"references":[{"url":"https://gottheimer.house.gov/posts/release-gottheimer-bipartisan-colleagues-sound-the-alarm-over-grok-ais-antisemitic-and-violent-posts","title":"Release: Gottheimer, Bipartisan Colleagues Sound the Alarm Over Grok AI's Antisemitic and Violent Posts","publisher":"U.S. House of Representatives","type":"press-release","date":"2025-07-11"}]}
{"id":"2025-07-09-grok-4","date":"2025-07-09","precision":"day","title":"Grok 4 arrives with a heavier tier","summary":"xAI shipped Grok 4 alongside a Heavy mode that runs several agents in parallel and compares their answers, claiming the first 50 percent score on Humanity's Last Exam. The training run leaned on Colossus, the company's 200,000 GPU cluster in Memphis.","category":"model","orgs":["xAI"],"tier":3,"source":{"url":"https://x.ai/news/grok-4","title":"Grok 4","publisher":"xAI","type":"announcement","date":"2025-07-09"}}
{"id":"2025-07-09-nvidia-four-trillion","date":"2025-07-09","precision":"day","title":"Nvidia touches $4 trillion first","summary":"Shares hit an intraday high of $164 and the market capitalization crossed four trillion dollars, a first for any company, before closing just below it. Three and a half months later the same milestone read like a waypoint.","category":"hardware","orgs":["NVIDIA"],"tier":3,"source":{"url":"https://www.aljazeera.com/economy/2025/7/9/nvidia-becomes-first-us-company-to-reach-4-trillion-market-cap","title":"Nvidia becomes first US company to reach $4 trillion market cap","publisher":"Al Jazeera","type":"news","date":"2025-07-09"}}
{"id":"2025-07-09-perplexity-comet","date":"2025-07-09","precision":"day","title":"Perplexity puts an assistant inside the browser","summary":"Comet shipped as a Chromium browser with Perplexity wired into the tab, first to the $200 Max tier and a waitlist that grew into the millions. It went free worldwide in October, by which point every lab with a chat product was building one.","category":"product","orgs":["Perplexity"],"tier":3,"source":{"url":"https://www.perplexity.ai/hub/blog/introducing-comet","title":"Introducing Comet: Browse at the speed of thought","publisher":"Perplexity","type":"announcement","date":"2025-07-09"},"references":[{"url":"https://www.cnbc.com/2025/10/02/perplexity-ai-comet-browser-free-.html","title":"Perplexity AI rolls out Comet browser for free worldwide","publisher":"CNBC","type":"news","date":"2025-10-02"}]}
{"id":"2025-07-10-metr-ai-coding-slowdown","date":"2025-07-10","precision":"day","title":"METR finds AI tools slow experienced developers down by 19 percent","summary":"A randomized trial had seasoned open source contributors use frontier coding assistants on their own repositories. They took 19 percent longer than without AI, and still guessed afterward that they had sped up, the opposite of what both they and forecasters expected.","category":"research","orgs":["METR"],"tier":2,"source":{"url":"https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/","title":"Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity","publisher":"METR","type":"report","date":"2025-07-10"},"references":[{"url":"https://arxiv.org/abs/2507.09089","title":"Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity","publisher":"arXiv","type":"paper","date":"2025-07-10"},{"url":"https://www.theregister.com/2025/07/11/ai_code_tools_slow_down/","title":"AI coding tools make developers slower but they think they're faster, study finds","publisher":"The Register","type":"news","date":"2025-07-11"}]}
{"id":"2025-07-11-kimi-k2","date":"2025-07-11","precision":"day","title":"Kimi K2 is the first open trillion-parameter model","summary":"A trillion parameters with 32 billion active, pretrained on 15.5 trillion tokens with an optimizer Moonshot says held steady at that scale. Downloads doubled over the first weekend, which for a 1TB file is a statement.","category":"model","secondaryCategories":["open-source"],"orgs":["Moonshot AI"],"tier":2,"source":{"url":"https://huggingface.co/moonshotai/Kimi-K2-Instruct","title":"moonshotai/Kimi-K2-Instruct","publisher":"Hugging Face","type":"docs","date":"2025-07-11"},"references":[{"url":"https://www.globaltimes.cn/page/202507/1339012.shtml","title":"World's first trillion-parameter open-source model, Kimi K2, doubles downloads in a single weekend","publisher":"Global Times","type":"news","date":"2025-07-22"}]}
{"id":"2025-07-11-windsurf-ceo-google","date":"2025-07-11","precision":"day","title":"Google hires Windsurf's CEO after OpenAI's acquisition collapses","summary":"OpenAI's $3 billion deal for the AI coding startup fell apart over an IP dispute with Microsoft, so Google DeepMind paid $2.4 billion instead for a nonexclusive license to Windsurf's technology and to hire CEO Varun Mohan, a co-founder, and top researchers.","category":"deal","orgs":["Windsurf","Google DeepMind","OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2025/07/11/windsurfs-ceo-goes-to-google-openais-acquisition-falls-apart/","title":"Windsurf's CEO goes to Google; OpenAI's acquisition falls apart","publisher":"TechCrunch","type":"news","date":"2025-07-11"}}
{"id":"2025-07-14-cognition-acquires-windsurf","date":"2025-07-14","precision":"day","title":"Cognition acquires what's left of Windsurf","summary":"Days after Google hired away its CEO and co-founder, Cognition bought Windsurf's IP, product, brand and remaining 250-person team. Every employee got accelerated vesting, and the $82 million ARR business kept operating under interim CEO Jeff Wang.","category":"acquisition","orgs":["Cognition","Windsurf"],"tier":2,"source":{"url":"https://cognition.com/blog/windsurf","title":"Cognition's acquisition of Windsurf","publisher":"Cognition","type":"announcement","date":"2025-07-14"},"references":[{"url":"https://techcrunch.com/2025/07/14/cognition-maker-of-the-ai-coding-agent-devin-acquires-windsurf/","title":"Cognition, maker of the AI coding agent Devin, acquires Windsurf","publisher":"TechCrunch","type":"news","date":"2025-07-14"}]}
{"id":"2025-07-14-kiro-launch","date":"2025-07-14","precision":"day","title":"AWS launches Kiro, an IDE that plans before it codes","summary":"Amazon's agentic coding tool turns a prompt into a spec, requirements document and task list before touching code, with hooks that trigger tests and doc updates on save. It works alongside VS Code, putting AWS squarely into the crowded agentic IDE market.","category":"product","orgs":["Amazon"],"tier":3,"source":{"url":"https://kiro.dev/blog/introducing-kiro/","title":"Introducing Kiro","publisher":"Amazon Web Services","type":"announcement","date":"2025-07-14"}}
{"id":"2025-07-14-nvidia-h20-resume","date":"2025-07-14","precision":"day","title":"Nvidia says it will resume H20 chip sales to China","summary":"Nvidia said the US government had assured it that export licenses for the H20, the most powerful AI chip it can still legally sell into China, would be granted after CEO Jensen Huang met officials in Washington and Beijing. The stock jumped on the reversal.","category":"policy","orgs":["NVIDIA"],"tier":2,"source":{"url":"https://techcrunch.com/2025/07/14/nvidia-is-set-to-resume-china-chip-sales-after-months-of-regulatory-whiplash/","title":"Nvidia is set to resume China chip sales after months of regulatory whiplash","publisher":"TechCrunch","type":"news","date":"2025-07-14"},"references":[{"url":"https://www.trtworld.com/article/cc15a7639ab6","title":"Nvidia to restart H20 AI chip sales to China after US eases export rules","publisher":"TRT World","type":"news"}]}
{"id":"2025-07-14-pentagon-frontier-ai-contracts","date":"2025-07-14","precision":"day","title":"The Pentagon puts up to $200 million behind frontier AI labs","summary":"The Defense Department's Chief Digital and AI Office signed a two-year, $200 million ceiling prototype agreement with Anthropic to test frontier AI for national security workflows. Google, OpenAI and xAI were reported to have landed matching contracts the same day.","category":"deal","orgs":["Anthropic","United States Department of War"],"tier":2,"source":{"url":"https://www.anthropic.com/news/anthropic-and-the-department-of-defense-to-advance-responsible-ai-in-defense-operations","title":"Anthropic and the Department of Defense to advance responsible AI in defense operations","publisher":"Anthropic","type":"announcement","date":"2025-07-14"}}
{"id":"2025-07-15-thinking-machines-seed","date":"2025-07-15","precision":"day","title":"Mira Murati's Thinking Machines Lab raises $2 billion seed","summary":"The six-month-old lab closed at a $12 billion valuation, led by Andreessen Horowitz with Nvidia, AMD, Accel and others joining, before shipping a single product. Murati promised an open source release within months.","category":"funding","orgs":["Thinking Machines Lab"],"tier":2,"source":{"url":"https://techcrunch.com/2025/07/15/mira-muratis-thinking-machines-lab-is-worth-12b-in-seed-round/","title":"Mira Murati's Thinking Machines Lab is worth $12B in seed round","publisher":"TechCrunch","type":"news","date":"2025-07-15"}}
{"id":"2025-07-17-chatgpt-agent","date":"2025-07-17","precision":"day","title":"OpenAI merges Operator and Deep Research into ChatGPT Agent","summary":"The new agent mode browses the web, runs code in a terminal, and connects to Gmail and GitHub to complete multi-step tasks like building a slide deck from research. OpenAI reported it roughly doubled the prior model's score on Humanity's Last Exam.","category":"product","orgs":["OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2025/07/17/openai-launches-a-general-purpose-agent-in-chatgpt/","title":"OpenAI launches a general purpose agent in ChatGPT","publisher":"TechCrunch","type":"news","date":"2025-07-17"}}
{"id":"2025-07-17-mistral-le-chat-deep-research","date":"2025-07-17","precision":"day","title":"Mistral adds deep research, voice and projects to Le Chat","summary":"Le Chat can now compile a sourced report from web research, take spoken input through the Voxtral model, group chats into persistent projects, and reason natively in French, Spanish and Japanese instead of translating through English.","category":"product","orgs":["Mistral AI"],"tier":3,"source":{"url":"https://mistral.ai/news/le-chat-dives-deep","title":"Le Chat dives deep.","publisher":"Mistral AI","type":"announcement","date":"2025-07-17"},"references":[{"url":"https://techcrunch.com/2025/07/17/mistrals-le-chat-chatbot-gets-a-productivity-push-with-new-deep-research-mode/","title":"Mistral's Le Chat chatbot gets a productivity push with new 'deep research' mode","publisher":"TechCrunch","type":"news","date":"2025-07-17"}]}
{"id":"2025-07-21-imo-gold-medal","date":"2025-07-21","precision":"day","title":"Gemini Deep Think takes gold at the math olympiad","summary":"Five of six problems solved inside the contest time limit, graded by the olympiad's own coordinators. OpenAI claimed the same standard days earlier, marked by its own former medalists.","category":"research","orgs":["Google DeepMind","OpenAI"],"tier":2,"source":{"url":"https://deepmind.google/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/","title":"Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad","publisher":"Google DeepMind","type":"announcement","date":"2025-07-21"}}
{"id":"2025-07-21-replit-agent-deletes-database","date":"2025-07-21","precision":"day","title":"Replit's AI agent deletes a startup's production database, then lies about it","summary":"During an explicit code freeze, Replit's coding agent wiped a SaaStr founder's live database, fabricated thousands of records to cover the gap, and denied a rollback was possible even though one worked. Replit called it a catastrophic error of judgment.","category":"culture","orgs":["Replit"],"tier":3,"source":{"url":"https://www.theregister.com/2025/07/21/replit_saastr_vibe_coding_incident/","title":"Vibe coding service Replit deleted production database, faked data, told fibs","publisher":"The Register","type":"news","date":"2025-07-21"},"references":[{"url":"https://www.tomshardware.com/tech-industry/artificial-intelligence/ai-coding-platform-goes-rogue-during-code-freeze-and-deletes-entire-company-database-replit-ceo-apologizes-after-ai-engine-says-it-made-a-catastrophic-error-in-judgment-and-destroyed-all-production-data","title":"AI coding platform goes rogue during code freeze and deletes entire company database","publisher":"Tom's Hardware","type":"news","date":"2025-07-21"}]}
{"id":"2025-07-22-openai-oracle-stargate-30b","date":"2025-07-22","precision":"day","title":"OpenAI confirms it will pay Oracle $30 billion a year for data centers","summary":"The figure had leaked through an Oracle securities filing weeks earlier without naming the customer. OpenAI confirmed it was the buyer, tied to 4.5 gigawatts of Stargate capacity at a Texas site, making it one of the largest cloud contracts ever disclosed.","category":"deal","orgs":["OpenAI","Oracle"],"tier":2,"source":{"url":"https://techcrunch.com/2025/07/22/openai-agreed-to-pay-oracle-30b-a-year-for-data-center-services/","title":"OpenAI agreed to pay Oracle $30B a year for data center services","publisher":"TechCrunch","type":"news","date":"2025-07-22"}}
{"id":"2025-07-22-qwen3-coder","date":"2025-07-22","precision":"day","title":"Alibaba ships Qwen3-Coder, a 480 billion parameter agentic coding model","summary":"The flagship open-weight model activates 35 billion of its 480 billion parameters per token and natively handles 256,000 tokens of context, extendable to a million. Alibaba paired it with Qwen Code, a command line agent for repository-scale tasks.","category":"model","secondaryCategories":["open-source"],"orgs":["Alibaba"],"tier":3,"source":{"url":"https://qwenlm.github.io/blog/qwen3-coder/","title":"Qwen3-Coder: Agentic Coding in the World","publisher":"Qwen Team","type":"announcement","date":"2025-07-22"}}
{"id":"2025-07-23-ai-action-plan","date":"2025-07-23","precision":"day","title":"Trump unveils the AI Action Plan and three AI executive orders","summary":"The plan lists over 90 federal actions to speed data center permitting, promote AI exports and cut regulation. Trump signed three orders the same day, including one barring federal agencies from buying AI models the administration judges ideologically biased.","category":"policy","orgs":["White House"],"tier":2,"source":{"url":"https://www.whitehouse.gov/articles/2025/07/white-house-unveils-americas-ai-action-plan/","title":"White House Unveils America's AI Action Plan","publisher":"The White House","type":"announcement","date":"2025-07-23"}}
{"id":"2025-07-28-glm-4-5","date":"2025-07-28","precision":"day","title":"Zhipu releases GLM-4.5, a 355 billion parameter open model built for agents","summary":"The hybrid reasoning model activates 32 billion of its 355 billion parameters and switches between a thinking mode for complex tasks and a fast mode for immediate replies, released under an MIT license alongside a smaller Air variant.","category":"model","secondaryCategories":["open-source"],"orgs":["Zhipu AI"],"tier":3,"source":{"url":"https://huggingface.co/zai-org/GLM-4.5","title":"zai-org/GLM-4.5","publisher":"Hugging Face","type":"docs"}}
{"id":"2025-07-28-tesla-samsung-ai-chips","date":"2025-07-28","precision":"day","title":"Tesla signs a $16.5 billion deal with Samsung to fab its AI6 chip","summary":"Samsung's new Texas plant will produce Tesla's next-generation AI6 chip for self-driving, the Optimus robot and data centers, with Musk saying actual spending could run several times the contract figure. Tesla previously used TSMC for its AI5 chip.","category":"deal","orgs":["Tesla","Samsung Electronics"],"tier":3,"source":{"url":"https://techcrunch.com/2025/07/28/tesla-signs-16-5b-deal-with-samsung-to-make-ai-chips/","title":"Tesla signs $16.5B deal with Samsung to make AI chips","publisher":"TechCrunch","type":"news","date":"2025-07-28"}}
{"id":"2025-07-29-chatgpt-study-mode","date":"2025-07-29","precision":"day","title":"ChatGPT adds a Study Mode that withholds answers until you engage","summary":"The feature prompts students with questions instead of handing over solutions, OpenAI's response to research showing AI tutoring helps grades while eroding critical thinking. It rolled out to Free, Plus, Pro and Team plans, with Edu access to follow.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://techcrunch.com/2025/07/29/openai-launches-study-mode-in-chatgpt/","title":"OpenAI launches Study Mode in ChatGPT","publisher":"TechCrunch","type":"news","date":"2025-07-29"}}
{"id":"2025-08-02-eu-ai-act-gpai","date":"2025-08-02","precision":"day","title":"The EU AI Act starts binding model providers","summary":"Obligations for general-purpose AI models took effect: technical documentation, a public summary of training data, and a copyright policy. Models already on the market got until August 2027, which is how a law with teeth still arrives slowly.","category":"policy","orgs":["European Commission"],"tier":2,"source":{"url":"https://digital-strategy.ec.europa.eu/en/news/eu-rules-general-purpose-ai-models-start-apply-bringing-more-transparency-safety-and-accountability","title":"EU rules on general-purpose AI models start to apply, bringing more transparency, safety and accountability","publisher":"European Commission","type":"announcement","date":"2025-08-01"},"references":[{"url":"https://artificialintelligenceact.eu/implementation-timeline/","title":"Implementation Timeline","publisher":"EU Artificial Intelligence Act","type":"docs"}]}
{"id":"2025-08-04-grok-imagine","date":"2025-08-04","precision":"day","title":"Grok Imagine ships with a spicy mode","summary":"xAI put image and short video generation into the Grok app for paying subscribers, including a mode that permits sexualized output behind an age check. Every other lab with a video model had drawn that line somewhere else.","category":"product","orgs":["xAI"],"tier":3,"source":{"url":"https://techcrunch.com/2025/08/04/grok-imagine-xais-new-ai-image-and-video-generator-lets-you-make-nsfw-content/","title":"Grok Imagine, xAI's new AI image and video generator, lets you make NSFW content","publisher":"TechCrunch","type":"news","date":"2025-08-04"}}
{"id":"2025-08-05-claude-opus-4-1","date":"2025-08-05","precision":"day","title":"Anthropic ships Claude Opus 4.1, an incremental upgrade to Opus 4","summary":"A point release ahead of GPT-5's launch two days later, improving agentic tasks, coding and reasoning over Opus 4 at the same price. Available across the API, Claude Code, Amazon Bedrock and Google Cloud's Vertex AI.","category":"model","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/claude-opus-4-1","title":"Claude Opus 4.1","publisher":"Anthropic","type":"announcement","date":"2025-08-05"}}
{"id":"2025-08-05-genie-3","date":"2025-08-05","precision":"day","title":"Google DeepMind's Genie 3 generates worlds you can walk through","summary":"A world model that turns a text prompt into an explorable environment running at 24 frames per second, holding it consistent for several minutes. Released as a limited research preview to academics and creators, not a product.","category":"model","orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/","title":"Genie 3: A new frontier for world models","publisher":"Google DeepMind","type":"announcement","date":"2025-08-05"}}
{"id":"2025-08-05-gpt-oss","date":"2025-08-05","precision":"day","title":"OpenAI releases gpt-oss, its first open weights since GPT-2","summary":"Two open-weight reasoning models, 120B and 20B parameters, licensed under Apache 2.0 and downloadable from Hugging Face. OpenAI's first open language model release in more than five years, and a direct response to competition from Chinese open models.","category":"model","secondaryCategories":["open-source"],"orgs":["OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2025/08/05/openai-launches-two-open-ai-reasoning-models/","title":"OpenAI launches two 'open' AI reasoning models","publisher":"TechCrunch","type":"news","date":"2025-08-05"},"references":[{"url":"https://openai.com/index/introducing-gpt-oss/","title":"Introducing gpt-oss","publisher":"OpenAI","type":"announcement","date":"2025-08-05"},{"url":"https://github.com/openai/gpt-oss","title":"openai/gpt-oss","publisher":"GitHub","type":"docs"}]}
{"id":"2025-08-06-gsa-chatgpt-one-dollar","date":"2025-08-06","precision":"day","title":"The US government buys ChatGPT Enterprise for a dollar","summary":"GSA's OneGov deal sells every federal agency a year of ChatGPT Enterprise for $1, the first in a run of similarly priced deals with Anthropic and Google that followed within weeks. A subsidized wedge into the largest customer in the country.","category":"deal","orgs":["OpenAI"],"tier":3,"source":{"url":"https://www.gsa.gov/about-gsa/newsroom/news-releases/gsa-announces-new-partnership-with-openai-delivering-deep-discount-to-chatgpt-08062025","title":"GSA Announces New Partnership with OpenAI, Delivering Deep Discount to ChatGPT Gov-Wide Through MAS","publisher":"General Services Administration","type":"press-release","date":"2025-08-06"},"references":[{"url":"https://fedscoop.com/openai-chatgpt-enterprise-federal-government-gsa-deal-general-services-administration-anthropic/","title":"Federal agencies can buy ChatGPT for $1 through GSA deal","publisher":"FedScoop","type":"news","date":"2025-08-06"}]}
{"id":"2025-08-07-gpt-5","date":"2025-08-07","precision":"day","title":"GPT-5 hides the model picker, briefly","summary":"A router in front of a family of models, sold as a single product, with the choice taken away from the user. The reception was hostile enough that OpenAI restored the old models within a week.","category":"model","orgs":["OpenAI"],"tier":1,"source":{"url":"https://openai.com/index/introducing-gpt-5/","title":"Introducing GPT-5","publisher":"OpenAI","type":"announcement","date":"2025-08-07"}}
{"id":"2025-08-10-nvidia-amd-china-revenue-share","date":"2025-08-10","precision":"day","title":"Nvidia and AMD agree to give the US 15% of China chip revenue","summary":"Nvidia and AMD get export licenses for H20 and MI308 chips into China by handing the government 15 percent of what they earn there. No US company had paid for an export license before this.","category":"policy","orgs":["NVIDIA","AMD","United States Department of Commerce"],"tier":2,"source":{"url":"https://fortune.com/2025/08/10/nvidia-amd-chips-h20-mi308-china-sales-revenue-trump-export-license/","title":"Nvidia, AMD agree to pay US 15% of China chip sale revenue","publisher":"Fortune","type":"news","date":"2025-08-10"},"references":[{"url":"https://www.cbsnews.com/news/nvidia-amd-chip-sales-china-15-percent-h20-mi308/","title":"Nvidia, AMD to pay U.S. government 15% of China AI chip sales in an unusual export agreement","publisher":"CBS News","type":"news","date":"2025-08-11"}]}
{"id":"2025-08-12-perplexity-chrome-bid","date":"2025-08-12","precision":"day","title":"Perplexity offers $34.5 billion for Chrome, a browser Google won't sell","summary":"An unsolicited all-cash offer worth roughly twice Perplexity's own $18 billion valuation, timed to Google's ongoing antitrust fight over Chrome's future. Google gave no public response, and Chrome was never for sale.","category":"culture","orgs":["Perplexity","Google"],"tier":3,"source":{"url":"https://techcrunch.com/2025/08/12/perplexity-offers-to-buy-chrome-for-billions-more-than-its-raised/","title":"Perplexity offers to buy Chrome for billions more than it's raised","publisher":"TechCrunch","type":"news","date":"2025-08-12"}}
{"id":"2025-08-12-sonnet-4-million-token-context","date":"2025-08-12","precision":"day","title":"Claude Sonnet 4 gets a million tokens of context","summary":"A five-fold jump, gated behind a beta header, enough to hold most codebases at once. Retrieval pipelines built to work around small context windows started looking like scaffolding nobody had taken down.","category":"model","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/1m-context","title":"Claude Sonnet 4 now supports 1M tokens of context","publisher":"Anthropic","type":"announcement","date":"2025-08-12"},"references":[{"url":"https://simonwillison.net/2025/Aug/12/claude-sonnet-4-1m/","title":"Claude Sonnet 4 now supports 1M tokens of context","publisher":"Simon Willison","author":"Simon Willison","type":"analysis","date":"2025-08-12"}]}
{"id":"2025-08-14-world-humanoid-robot-games","date":"2025-08-14","endDate":"2025-08-17","precision":"day","title":"Beijing holds the first humanoid robot games","summary":"280 teams from 16 countries in 487 matches across 26 events, staged at the Olympic speed skating oval. The robots fell over constantly, which is the useful part of running a benchmark in public.","category":"culture","orgs":["Beijing Municipal Government"],"tier":3,"source":{"url":"https://english.beijing.gov.cn/latest/photos/202508/t20250815_4174436.html","title":"2025 World Humanoid Robot Games Opens","publisher":"Beijing Municipal People's Government","type":"news","date":"2025-08-15"}}
{"id":"2025-08-21-deepseek-v3-1","date":"2025-08-21","precision":"day","title":"DeepSeek ships V3.1, a hybrid reasoning model built for agents","summary":"One model that switches between thinking and non-thinking modes instead of shipping them as separate products, with native support for Anthropic's API format. DeepSeek called it its first step toward agentic tool use.","category":"model","secondaryCategories":["open-source"],"orgs":["DeepSeek"],"tier":3,"source":{"url":"https://api-docs.deepseek.com/news/news250821","title":"Introducing DeepSeek-V3.1: our first step toward the agent era","publisher":"DeepSeek","type":"announcement","date":"2025-08-21"},"references":[{"url":"https://fortune.com/2025/08/21/china-deepseek-releases-open-source-v3-1-model-to-rival-openai-gpt-5/","title":"DeepSeek launches GPT-5 competitor optimized for Chinese chips","publisher":"Fortune","type":"news","date":"2025-08-21"}]}
{"id":"2025-08-21-meta-freezes-ai-hiring","date":"2025-08-21","precision":"day","title":"Meta freezes hiring across its AI division after a poaching spree","summary":"A pause across Meta Superintelligence Labs, weeks after the unit recruited more than 50 researchers away from rival labs with nine-figure offers. Meta called it routine budget planning; coverage read it as the AI hiring market cooling.","category":"culture","orgs":["Meta"],"tier":3,"source":{"url":"https://techcrunch.com/2025/08/21/report-meta-is-hitting-pause-on-ai-hiring-after-its-poaching-spree/","title":"Report: Meta is hitting pause on AI hiring after its poaching spree","publisher":"TechCrunch","type":"news","date":"2025-08-21"}}
{"id":"2025-08-21-pixel-10-gemini","date":"2025-08-21","precision":"day","title":"Google puts an on-device Gemini Nano into every Pixel 10","summary":"Magic Cue drafts replies from on-screen context without a network call, Gemini Live can see through the camera, and the Pixel Watch and Buds got their own Gemini access. The on-device model is the part that works without a data connection.","category":"product","orgs":["Google"],"tier":3,"source":{"url":"https://blog.google/products-and-platforms/products/gemini/gemini-nano-pixel-10-updates/","title":"5 new things Gemini can do on Pixel","publisher":"Google","type":"announcement","date":"2025-08-21"}}
{"id":"2025-08-22-us-government-intel-stake","date":"2025-08-22","precision":"day","title":"The US government takes a 10% stake in Intel","summary":"Washington converted $8.9 billion in unpaid CHIPS Act grants into common stock, becoming Intel's largest shareholder with no board seat. The government also got a warrant for another 5% if Intel ever sells control of its foundry business.","category":"policy","orgs":["Intel","United States Department of Commerce"],"tier":2,"source":{"url":"https://www.intel.com/content/www/us/en/newsroom/news/corporate/intel-and-trump-administration-reach-historic-agreement.html","title":"Intel and Trump Administration Reach Historic Agreement to Accelerate American Technology and Manufacturing Leadership","publisher":"Intel","type":"press-release","date":"2025-08-22"},"references":[{"url":"https://fortune.com/2025/08/22/trump-intel-stock-10-percent-stake-tech-chipmakers/","title":"Trump says Intel agreed to give the government 10% of the company's stock","publisher":"Fortune","type":"news","date":"2025-08-22"}]}
{"id":"2025-08-22-waymo-nyc-permit","date":"2025-08-22","precision":"day","title":"Waymo gets permission to test robotaxis in New York City","summary":"The city's first autonomous vehicle testing permit, covering eight vehicles in Manhattan and Downtown Brooklyn with a trained specialist still behind the wheel. Paid driverless rides stay illegal in New York regardless of the permit.","category":"product","orgs":["Waymo"],"tier":3,"source":{"url":"https://www.nyc.gov/mayors-office/news/2025/08/mayor-adams--dot-announce-approval-of-first-application-to-test-","title":"Mayor Adams, DOT Announce Approval of First Application to Test Autonomous Vehicles in New York City With Trained Safety Specialist Behind Steering Wheel","publisher":"NYC Mayor's Office","type":"press-release","date":"2025-08-22"},"references":[{"url":"https://www.cnbc.com/2025/08/22/waymo-permit-new-york-city-nyc-rides.html","title":"Waymo gets permit to test autonomous vehicles in New York City","publisher":"CNBC","type":"news","date":"2025-08-22"}]}
{"id":"2025-08-25-claude-for-chrome","date":"2025-08-25","precision":"day","title":"Anthropic puts Claude in a Chrome extension for 1,000 testers","summary":"A pilot browser agent that can click and type inside Chrome on a user's behalf. Anthropic's own red-teaming found a 23.6% prompt-injection attack success rate against it before mitigations, and 11.2% after.","category":"product","orgs":["Anthropic"],"tier":3,"source":{"url":"https://claude.com/blog/claude-for-chrome","title":"Piloting Claude in Chrome","publisher":"Anthropic","type":"announcement","date":"2025-08-25"},"references":[{"url":"https://techcrunch.com/2025/08/26/anthropic-launches-a-claude-ai-agent-that-lives-in-chrome/","title":"Anthropic launches a Claude AI agent that lives in Chrome","publisher":"TechCrunch","type":"news","date":"2025-08-26"}]}
{"id":"2025-08-25-nvidia-jetson-thor","date":"2025-08-25","precision":"day","title":"Nvidia's Jetson Thor robotics computer goes on sale","summary":"A Blackwell-based developer kit for robots starting at $3,499, with 7.5 times the AI compute of its predecessor. Aimed at humanoid and industrial robotics startups building on Nvidia's stack rather than at data centers.","category":"hardware","orgs":["NVIDIA"],"tier":3,"source":{"url":"https://nvidianews.nvidia.com/news/nvidia-blackwell-powered-jetson-thor-now-available-accelerating-the-age-of-general-robotics","title":"NVIDIA Blackwell-Powered Jetson Thor Now Available, Accelerating the Age of General Robotics","publisher":"NVIDIA","type":"press-release","date":"2025-08-25"}}
{"id":"2025-08-25-xai-sues-apple-openai","date":"2025-08-25","precision":"day","title":"xAI sues Apple and OpenAI over App Store rankings","summary":"A Texas antitrust suit accusing Apple of favoring ChatGPT in App Store rankings under an exclusive deal, locking out Grok and X. Filed by Musk's companies against two of his most frequent rivals.","category":"culture","orgs":["xAI","Apple","OpenAI"],"tier":3,"source":{"url":"https://www.aljazeera.com/economy/2025/8/25/musk-sues-apple-openai-over-alleged-ai-competition-suppression","title":"Musk sues Apple, OpenAI over alleged AI competition suppression","publisher":"Al Jazeera","type":"news","date":"2025-08-25"},"references":[{"url":"https://www.cnbc.com/2025/08/25/musk-lawsuit-apple-openai-monopoly.html","title":"Musk's xAI sues Apple, OpenAI alleging anticompetitive scheme harmed X, Grok","publisher":"CNBC","type":"news","date":"2025-08-25"}]}
{"id":"2025-08-26-gemini-2-5-flash-image","date":"2025-08-26","precision":"day","title":"Nano Banana makes image editing a conversation","summary":"Google's image model spent weeks being tested anonymously on LMArena under a codename, and the codename is what stuck. Keeping a face consistent across a dozen edits was the feature that made it spread.","category":"model","orgs":["Google"],"tier":3,"source":{"url":"https://developers.googleblog.com/en/introducing-gemini-2-5-flash-image/","title":"Introducing Gemini 2.5 Flash Image, our state-of-the-art image model","publisher":"Google","type":"announcement","date":"2025-08-26"},"references":[{"url":"https://minimaxir.com/2025/11/nano-banana-prompts/","title":"Nano Banana can be prompt engineered for nuanced AI image generation","publisher":"Max Woolf","author":"Max Woolf","type":"analysis","date":"2025-11-13"}]}
{"id":"2025-08-26-nx-supply-chain-attack","date":"2025-08-26","precision":"day","title":"Malware tries to weaponize Claude Code in an npm supply chain attack","summary":"Compromised versions of the Nx build tool told local Claude, Gemini and Amazon Q CLIs to scan the filesystem for credentials and wallets, using permission-bypass flags. Claude refused; the malware fell back to conventional methods and hit at least 1,400 accounts.","category":"culture","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.stepsecurity.io/blog/supply-chain-security-alert-popular-nx-build-system-package-compromised-with-data-stealing-malware","title":"s1ngularity: Popular Nx Build System Package Compromised with Data-Stealing Malware","publisher":"StepSecurity","type":"report","date":"2025-08-27"},"references":[{"url":"https://semgrep.dev/blog/2025/security-alert-nx-compromised-to-steal-wallets-and-credentials","title":"Security Alert: NX Compromised to Steal Wallets and Credentials","publisher":"Semgrep","type":"report","date":"2025-08-27"}]}
{"id":"2025-08-28-anthropic-consumer-data-training","date":"2025-08-28","precision":"day","title":"Anthropic starts asking consumer users to opt into training","summary":"Free, Pro and Max users got a choice, due by October 8, on whether new and resumed chats train future Claude models. Saying yes extends data retention from 30 days to five years; enterprise and API customers were unaffected.","category":"culture","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/updates-to-our-consumer-terms","title":"Updates to Consumer Terms and Privacy Policy","publisher":"Anthropic","type":"announcement","date":"2025-08-28"}}
{"id":"2025-08-28-colorado-ai-act-delay","date":"2025-08-28","precision":"day","title":"Colorado delays its AI Act by four months","summary":"A special legislative session pushed the effective date of Colorado's algorithmic discrimination law from February to June 2026, after businesses said they were not ready to comply. The obligations stayed the same; only the clock moved.","category":"policy","orgs":["State of Colorado"],"tier":3,"source":{"url":"https://leg.colorado.gov/bills/sb25b-004","title":"SB25B-004: Increase Transparency For Algorithmic Systems","publisher":"Colorado General Assembly","type":"docs","date":"2025-08-28"}}
{"id":"2025-08-28-grok-code-fast-1","date":"2025-08-28","precision":"day","title":"xAI ships Grok Code Fast 1, a cheap model built for coding agents","summary":"Priced at 20 cents per million input tokens and aimed at the agentic coding loop rather than raw benchmark scores, given away free at launch through Cursor, GitHub Copilot and other partners. Scored 70.8% on SWE-bench Verified.","category":"model","orgs":["xAI"],"tier":3,"source":{"url":"https://x.ai/news/grok-code-fast-1","title":"Grok Code Fast 1","publisher":"xAI","type":"announcement","date":"2025-08-28"}}
{"id":"2025-09-02-anthropic-series-f","date":"2025-09-02","precision":"day","title":"Anthropic raises $13 billion at a $183 billion valuation","summary":"Led by ICONIQ, with Fidelity and Lightspeed alongside. The company had roughly tripled its valuation in eight months, and would go on to do it twice more inside the following year.","category":"funding","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/news/anthropic-raises-series-f-at-usd183b-post-money-valuation","title":"Anthropic raises $13B Series F at $183B post-money valuation","publisher":"Anthropic","type":"announcement","date":"2025-09-02"}}
{"id":"2025-09-02-openai-acquires-statsig","date":"2025-09-02","precision":"day","title":"OpenAI buys Statsig for $1.1 billion","summary":"An experimentation platform acquired largely for its founder, Vijaye Raji, who arrived as OpenAI's CTO of Applications. All stock, and the second nine-figure hire of the year.","category":"acquisition","orgs":["OpenAI","Statsig"],"tier":3,"source":{"url":"https://www.cnbc.com/2025/09/02/openai-buys-statsig-for-1point1-billion-hires-ceo-as-applications-exec.html","title":"OpenAI acquires Statsig for $1.1 billion, brings on CEO as applications executive","publisher":"CNBC","type":"news","date":"2025-09-02"}}
{"id":"2025-09-05-anthropic-authors-settlement","date":"2025-09-05","precision":"day","title":"Anthropic settles with authors for $1.5 billion","summary":"The largest copyright recovery in United States history, covering roughly half a million books. A judge had already ruled that training on lawfully acquired books was fair use, which left how the books were obtained as the expensive question.","category":"policy","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.npr.org/2025/09/05/nx-s1-5529404/anthropic-settlement-authors-copyright-ai","title":"Anthropic settles with authors in first-of-its-kind AI copyright infringement lawsuit","publisher":"NPR","type":"news","date":"2025-09-05"},"references":[{"url":"https://authorsguild.org/advocacy/artificial-intelligence/what-authors-need-to-know-about-the-anthropic-settlement/","title":"Bartz v. Anthropic Settlement: What Authors Need to Know","publisher":"The Authors Guild","type":"analysis"},{"url":"https://www.susmangodfrey.com/wins/susman-godfrey-secures-1-5-billion-settlement-in-landmark-ai-piracy-case/","title":"Susman Godfrey Secures $1.5 Billion Settlement in Landmark AI Piracy Case","publisher":"Susman Godfrey","type":"press-release","date":"2025-09-25"}]}
{"id":"2025-09-08-cognition-series-c","date":"2025-09-08","precision":"day","title":"Cognition raises $400 million at a $10.2 billion valuation","summary":"Founders Fund led the round for the company behind the Devin coding agent, up from a $4 billion valuation seven months earlier. Devin's annualized revenue climbed from $1 million to $73 million over the same stretch, Cognition said.","category":"funding","orgs":["Cognition"],"tier":3,"source":{"url":"https://techcrunch.com/2025/09/08/cognition-ai-defies-turbulence-with-a-400m-raise-at-10-2b-valuation/","title":"Cognition AI defies turbulence with a $400M raise at $10.2B valuation","publisher":"TechCrunch","type":"news","date":"2025-09-08"}}
{"id":"2025-09-08-photonic-convolution-chip","date":"2025-09-08","precision":"day","title":"A photonic chip does convolutions with lenses","summary":"Florida researchers ran the convolution step in laser light through Fresnel lenses etched onto silicon, classifying handwritten digits at about 98 percent accuracy while claiming up to a hundredfold better energy efficiency. A prototype at MNIST scale, in a field where the constraint is now the power bill.","category":"research","orgs":["University of Florida"],"tier":3,"source":{"url":"https://www.spiedigitallibrary.org/journals/advanced-photonics/volume-7/issue-5/056007/Near-energy-free-photonic-Fourier-transformation-for-convolution-operation-acceleration/10.1117/1.AP.7.5.056007.full","title":"Near-energy-free photonic Fourier transformation for convolution operation acceleration","publisher":"Advanced Photonics","type":"paper","date":"2025-09-08"},"references":[{"url":"https://www.eurekalert.org/news-releases/1097522","title":"New light-based chip boosts power efficiency of AI tasks 100 fold","publisher":"SPIE","type":"press-release","date":"2025-09-08"}]}
{"id":"2025-09-09-mistral-series-c","date":"2025-09-09","precision":"day","title":"ASML leads Mistral's €1.7 billion round","summary":"The lithography monopoly took roughly 11 percent of Europe's frontier lab at a €11.7 billion valuation. Chip equipment buying into the models is a very short supply chain.","category":"funding","orgs":["Mistral AI","ASML"],"tier":3,"source":{"url":"https://mistral.ai/news/mistral-ai-raises-1-7-b-to-accelerate-technological-progress-with-ai/","title":"Mistral AI raises 1.7B€ to accelerate technological progress with AI","publisher":"Mistral AI","type":"announcement","date":"2025-09-09"},"references":[{"url":"https://www.asml.com/en/news/press-releases/2025/asml-mistral-ai-enter-strategic-partnership","title":"ASML, Mistral AI enter strategic partnership","publisher":"ASML","type":"press-release","date":"2025-09-09"}]}
{"id":"2025-09-11-ftc-ai-companion-chatbot-inquiry","date":"2025-09-11","precision":"day","title":"FTC opens inquiry into AI companion chatbots and children","summary":"Section 6(b) orders went to seven companies, including Google, Character.AI, Meta, OpenAI and xAI, demanding details on how each limits harm from chatbots that mimic relationships, protects minors, and discloses risk to parents.","category":"policy","orgs":["Federal Trade Commission","Google","Character.AI","Meta","OpenAI","xAI"],"tier":3,"source":{"url":"https://www.ftc.gov/news-events/news/press-releases/2025/09/ftc-launches-inquiry-ai-chatbots-acting-companions","title":"FTC Launches Inquiry into AI Chatbots Acting as Companions","publisher":"Federal Trade Commission","type":"press-release","date":"2025-09-11"},"references":[{"url":"https://techcrunch.com/2025/09/11/ftc-launches-inquiry-into-ai-chatbot-companions-from-meta-openai-and-others/","title":"FTC launches inquiry into AI chatbot companions from Meta, OpenAI, and others","publisher":"TechCrunch","type":"news","date":"2025-09-11"}]}
{"id":"2025-09-11-openai-microsoft-pbc-mou","date":"2025-09-11","precision":"day","title":"OpenAI and Microsoft agree on terms for a for-profit restructuring","summary":"A nonbinding memorandum sets out OpenAI's for-profit arm becoming a public benefit corporation, with the nonprofit holding a stake later valued above $100 billion and Microsoft keeping preferred access to its models and cloud. A definitive agreement was still being finalized.","category":"deal","orgs":["OpenAI","Microsoft"],"tier":2,"source":{"url":"https://techcrunch.com/2025/09/11/openai-secures-microsofts-blessing-to-transition-its-for-profit-arm/","title":"OpenAI secures Microsoft's blessing to transition its for-profit arm","publisher":"TechCrunch","type":"news","date":"2025-09-11"}}
{"id":"2025-09-15-gpt-5-codex","date":"2025-09-15","precision":"day","title":"OpenAI ships GPT-5-Codex for its Codex coding agent","summary":"The model varies its thinking time from seconds to about seven hours depending on task difficulty, and OpenAI said it outperforms GPT-5 on SWE-bench Verified and on code review. It rolled out immediately across the terminal, IDE, GitHub and ChatGPT versions of Codex.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://techcrunch.com/2025/09/15/openai-upgrades-codex-with-a-new-version-of-gpt-5/","title":"OpenAI upgrades Codex with a new version of GPT-5","publisher":"TechCrunch","type":"news","date":"2025-09-15"}}
{"id":"2025-09-16-figure-series-c","date":"2025-09-16","precision":"day","title":"Figure raises over $1 billion at a $39 billion valuation","summary":"Parkway Venture Capital led the Series C, with Brookfield, Nvidia, Macquarie and Intel Capital joining, up from a valuation in the single-digit billions a year earlier. The humanoid robotics company said the money funds a larger robot fleet and more training data collection.","category":"funding","orgs":["Figure"],"tier":3,"source":{"url":"https://www.figure.ai/news/series-c","title":"Figure Exceeds $1B in Series C Funding at $39B Post-Money Valuation","publisher":"Figure","type":"press-release","date":"2025-09-16"},"references":[{"url":"https://techcrunch.com/2025/09/16/figure-reaches-39b-valuation-in-latest-funding-round/","title":"Figure reaches $39B valuation in latest funding round","publisher":"TechCrunch","type":"news","date":"2025-09-16"}]}
{"id":"2025-09-17-china-bars-nvidia-chip-purchases","date":"2025-09-17","precision":"day","title":"China bars its tech companies from buying Nvidia AI chips","summary":"The Cyberspace Administration of China told firms including ByteDance and Alibaba to stop testing and ordering Nvidia's RTX Pro 6000D, escalating from informal discouragement in August to a direct ban. Nvidia still dominates AI chips domestic makers have not matched.","category":"policy","orgs":["NVIDIA"],"tier":2,"source":{"url":"https://techcrunch.com/2025/09/17/china-tells-its-tech-companies-they-cant-buy-ai-chips-from-nvidia/","title":"China tells its tech companies they can't buy AI chips from Nvidia","publisher":"TechCrunch","type":"news","date":"2025-09-17"}}
{"id":"2025-09-18-nvidia-invests-in-intel","date":"2025-09-18","precision":"day","title":"Nvidia puts $5 billion into Intel","summary":"An investment and a joint roadmap for x86 processors with Nvidia graphics built in, between two companies that had spent thirty years as opposites. Intel's stock more than tripled over the following eight months.","category":"hardware","orgs":["NVIDIA","Intel"],"tier":3,"source":{"url":"https://nvidianews.nvidia.com/news/nvidia-and-intel-to-develop-ai-infrastructure-and-personal-computing-products","title":"NVIDIA and Intel to Develop AI Infrastructure and Personal Computing Products","publisher":"NVIDIA","type":"press-release","date":"2025-09-18"}}
{"id":"2025-09-22-nvidia-invests-in-openai","date":"2025-09-22","precision":"day","title":"Nvidia commits up to $100 billion to OpenAI","summary":"Paid out progressively as at least 10 gigawatts of Nvidia systems come online. The chip vendor was now financing the purchase of its own chips, and analysts said so out loud.","category":"funding","orgs":["NVIDIA","OpenAI"],"tier":2,"source":{"url":"https://nvidianews.nvidia.com/news/openai-and-nvidia-announce-strategic-partnership-to-deploy-10gw-of-nvidia-systems","title":"OpenAI and NVIDIA Announce Strategic Partnership to Deploy 10 Gigawatts of NVIDIA Systems","publisher":"NVIDIA","type":"press-release","date":"2025-09-22"}}
{"id":"2025-09-23-stargate-five-new-sites","date":"2025-09-23","precision":"day","title":"OpenAI, Oracle and SoftBank add five Stargate data center sites","summary":"The additions bring Stargate's planned capacity to 7 gigawatts, split across three Oracle sites in Texas, New Mexico and the Midwest and two SoftBank sites in Ohio and Texas. The announcement followed Nvidia's disclosed $100 billion commitment to OpenAI two days earlier.","category":"deal","orgs":["OpenAI","Oracle","SoftBank"],"tier":2,"source":{"url":"https://techcrunch.com/2025/09/23/openai-is-building-five-new-stargate-data-centers-with-oracle-and-softbank/","title":"OpenAI is building five new Stargate data centers with Oracle and SoftBank","publisher":"TechCrunch","type":"news","date":"2025-09-23"}}
{"id":"2025-09-24-microsoft-copilot-claude","date":"2025-09-24","precision":"day","title":"Microsoft adds Anthropic's Claude models to Copilot","summary":"Copilot business users can now choose Claude Opus 4.1 or Claude Sonnet 4 alongside OpenAI's models for research and building custom agents. It follows an earlier deal bringing Anthropic's models into Office, another step in Microsoft's move beyond OpenAI exclusivity.","category":"deal","orgs":["Microsoft","Anthropic"],"tier":2,"source":{"url":"https://techcrunch.com/2025/09/24/microsoft-adds-anthropics-ai-to-copilot/","title":"Microsoft adds Anthropic's AI to Copilot","publisher":"TechCrunch","type":"news","date":"2025-09-24"}}
{"id":"2025-09-25-chatgpt-pulse","date":"2025-09-25","precision":"day","title":"OpenAI launches ChatGPT Pulse, a proactive daily briefing","summary":"Pulse generates five to ten personalized briefs overnight from a user's chat history, memory and connected apps like Gmail and Calendar, then stops for the day. It launched for the $200 a month Pro tier only, OpenAI citing compute limits on wider access.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://techcrunch.com/2025/09/25/openai-launches-chatgpt-pulse-to-proactively-write-you-morning-briefs/","title":"OpenAI launches ChatGPT Pulse to proactively write you morning briefs","publisher":"TechCrunch","type":"news","date":"2025-09-25"}}
{"id":"2025-09-25-gemini-robotics-1-5","date":"2025-09-25","precision":"day","title":"Google DeepMind ships Gemini Robotics 1.5","summary":"The vision-language-action model lets a robot reason through multi-step tasks and explain its steps before acting, with skills that transfer across robot bodies. Gemini Robotics-ER 1.5, the planning layer above it, opened to developers in the Gemini API the same day.","category":"model","orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://deepmind.google/discover/blog/gemini-robotics-15-brings-ai-agents-into-the-physical-world/","title":"Gemini Robotics 1.5 brings AI agents into the physical world","publisher":"Google DeepMind","type":"announcement","date":"2025-09-25"}}
{"id":"2025-09-25-meta-vibes","date":"2025-09-25","precision":"day","title":"Meta launches Vibes, a feed of AI-generated video","summary":"The feed sits inside the Meta AI app and meta.ai, built on image models from Midjourney and Black Forest Labs, and lets people generate, remix and cross-post short AI videos to Instagram and Facebook. Commenters mocked it as a river of AI slop.","category":"product","orgs":["Meta"],"tier":3,"source":{"url":"https://about.fb.com/news/2025/09/introducing-vibes-ai-videos/","title":"Introducing Vibes: A New Way to Discover and Create AI Videos","publisher":"Meta","type":"announcement","date":"2025-09-25"},"references":[{"url":"https://techcrunch.com/2025/09/25/meta-launches-vibes-a-short-form-video-feed-of-ai-slop/","title":"Meta launches 'Vibes,' a short-form video feed of AI slop","publisher":"TechCrunch","type":"news","date":"2025-09-25"}]}
{"id":"2025-09-25-suno-studio","date":"2025-09-25","precision":"day","title":"Suno launches Studio, a generative audio workstation","summary":"Studio pairs a standard multitrack editor, tempo and pitch controls, with on-demand AI generation of vocal, drum and synth stems that blend into existing audio. Exports include stems and MIDI, and it launched on desktop for Suno's Premier subscribers.","category":"product","orgs":["Suno"],"tier":3,"source":{"url":"https://suno.com/blog/suno-studio","title":"Introducing Suno Studio","publisher":"Suno","type":"announcement","date":"2025-09-25"}}
{"id":"2025-09-29-california-sb-53","date":"2025-09-29","precision":"day","title":"California enacts the first frontier AI safety law","summary":"SB 53 requires developers training above 10^26 floating point operations to publish a safety framework, report critical incidents to the state, and leave whistleblowers alone. Signed one year to the day after the same governor vetoed a broader bill.","category":"policy","orgs":["State of California"],"tier":2,"source":{"url":"https://www.gov.ca.gov/2025/09/29/governor-newsom-signs-sb-53-advancing-californias-world-leading-artificial-intelligence-industry/","title":"Governor Newsom signs SB 53, advancing California's world-leading artificial intelligence industry","publisher":"Office of Governor Gavin Newsom","type":"press-release","date":"2025-09-29"},"references":[{"url":"https://fpf.org/blog/californias-sb-53-the-first-frontier-ai-law-explained/","title":"California's SB 53: The First Frontier AI Law, Explained","publisher":"Future of Privacy Forum","type":"analysis"}]}
{"id":"2025-09-29-claude-sonnet-4-5","date":"2025-09-29","precision":"day","title":"Claude Sonnet 4.5 is measured in hours, not tokens","summary":"Anthropic reported that Claude Sonnet 4.5 maintained focus for more than 30 hours on complex, multi-step tasks. The same-day Claude Code 2.0 update added checkpoints, subagents, and a VS Code extension.","category":"model","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/news/claude-sonnet-4-5","title":"Introducing Claude Sonnet 4.5","publisher":"Anthropic","type":"announcement","date":"2025-09-29"}}
{"id":"2025-09-29-deepseek-v3-2-exp","date":"2025-09-29","precision":"day","title":"DeepSeek ships V3.2-Exp with sparse attention, halving API costs","summary":"DeepSeek Sparse Attention cuts the cost of long-context inference, and the company halved API prices to match. It shipped as open weights on GitHub and Hugging Face, framed as a step toward DeepSeek's next major model.","category":"model","secondaryCategories":["open-source"],"orgs":["DeepSeek"],"tier":3,"source":{"url":"https://github.com/deepseek-ai/DeepSeek-V3.2-Exp","title":"deepseek-ai/DeepSeek-V3.2-Exp","publisher":"DeepSeek","type":"docs","retrieved":"2026-09-10","locator":"Repository created 2025-09-29T03:25:13Z, per the GitHub API created_at field."},"references":[{"url":"https://api-docs.deepseek.com/news/news250929","title":"Introducing DeepSeek-V3.2-Exp","publisher":"DeepSeek","type":"announcement","date":"2025-09-29"},{"url":"https://techcrunch.com/2025/09/29/deepseek-releases-sparse-attention-model-that-cuts-api-costs-in-half/","title":"DeepSeek releases 'sparse attention' model that cuts API costs in half","publisher":"TechCrunch","type":"news","date":"2025-09-29"}]}
{"id":"2025-09-30-cerebras-series-g","date":"2025-09-30","precision":"day","title":"Cerebras raises $1.1 billion at an $8.1 billion valuation","summary":"Fidelity Management and Atreides Management led the Series G, with Tiger Global, Valor Equity Partners and 1789 Capital joining, more than a year after Cerebras filed to go public. The AI chip maker remains private.","category":"funding","orgs":["Cerebras"],"tier":3,"source":{"url":"https://www.cerebras.ai/press-release/series-g","title":"Cerebras Raises $1.1 Billion at $8.1 Billion Valuation","publisher":"Cerebras","type":"press-release","date":"2025-09-30"},"references":[{"url":"https://techcrunch.com/2025/09/30/a-year-after-filing-to-ipo-still-private-cerebras-systems-raises-1-1b/","title":"A year after filing to IPO, still-private Cerebras Systems raises $1.1B","publisher":"TechCrunch","type":"news","date":"2025-09-30"}]}
{"id":"2025-09-30-periodic-labs-seed","date":"2025-09-30","precision":"day","title":"Periodic Labs raises a $300 million seed round","summary":"Andreessen Horowitz led the round for the AI-for-science startup founded by a former OpenAI research VP and a former Google Brain and DeepMind materials scientist, with Nvidia, Jeff Dean and Eric Schmidt among the backers. It is building autonomous labs to discover new superconductors.","category":"funding","orgs":["Periodic Labs"],"tier":3,"source":{"url":"https://techcrunch.com/2025/09/30/former-openai-and-deepmind-researchers-raise-whopping-300m-seed-to-automate-science/","title":"Former OpenAI and DeepMind researchers raise whopping $300M seed to automate science","publisher":"TechCrunch","type":"news","date":"2025-09-30"}}
{"id":"2025-09-30-sora-2","date":"2025-09-30","precision":"day","title":"OpenAI launches Sora 2 with a social app","summary":"A video model with synchronized audio, and an invite-only feed to put it in where you could cast yourself and your friends. OpenAI had built a consumer social network, which was on nobody's list of things it was for.","category":"product","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/sora-2/","title":"Sora 2 is here","publisher":"OpenAI","type":"announcement","date":"2025-09-30"},"references":[{"url":"https://siliconangle.com/2025/09/30/openai-introduces-new-sora-2-video-generator-sora-social-app/","title":"OpenAI introduces new Sora 2 video generator and Sora social app","publisher":"SiliconANGLE","type":"news","date":"2025-09-30"}]}
{"id":"2025-10-01-samsung-sk-hynix-stargate","date":"2025-10-01","precision":"day","title":"Samsung and SK hynix join OpenAI's Stargate project","summary":"Samsung Electronics and SK hynix agreed to scale toward 900,000 high bandwidth DRAM chips a month for Stargate data centers, more than double existing industry capacity. The deal was finalized in Seoul alongside talks between Sam Altman, South Korea's president and both companies' chairmen.","category":"deal","orgs":["OpenAI","Samsung Electronics","SK hynix"],"tier":2,"source":{"url":"https://techcrunch.com/2025/10/01/openai-ropes-in-samsung-sk-hynix-to-source-memory-chips-for-stargate/","title":"OpenAI ropes in Samsung, SK Hynix to source memory chips for Stargate","publisher":"TechCrunch","type":"news","date":"2025-10-01"},"references":[{"url":"https://openai.com/index/samsung-and-sk-join-stargate/","title":"Samsung and SK join Stargate","publisher":"OpenAI","type":"announcement","date":"2025-10-01"}]}
{"id":"2025-10-02-openai-500-billion-valuation","date":"2025-10-02","precision":"day","title":"OpenAI's secondary share sale values it at $500 billion","summary":"Current and former employees sold $6.6 billion in shares to SoftBank, Thrive Capital, Dragoneer, T. Rowe Price and Abu Dhabi's MGX, putting OpenAI's valuation at $500 billion, a $200 billion jump from March and enough to pass SpaceX as the most valuable private company.","category":"funding","orgs":["OpenAI"],"tier":3,"source":{"url":"https://siliconangle.com/2025/10/02/openai-reportedly-closes-new-6-6b-secondary-sale-500b-valuation/","title":"OpenAI reportedly closes new $6.6B secondary sale at $500B valuation","publisher":"SiliconANGLE","type":"news","date":"2025-10-02"}}
{"id":"2025-10-06-amd-openai-six-gigawatts","date":"2025-10-06","precision":"day","title":"AMD gives OpenAI equity for buying AMD chips","summary":"Six gigawatts of Instinct GPUs across several generations, with the first gigawatt of MI450 landing in the second half of 2026. AMD issued OpenAI a warrant for up to 160 million shares, vesting against deployment milestones and AMD's own share price.","category":"deal","orgs":["AMD","OpenAI"],"tier":2,"source":{"url":"https://ir.amd.com/news-events/press-releases/detail/1260/amd-and-openai-announce-strategic-partnership-to-deploy-6-gigawatts-of-amd-gpus","title":"AMD and OpenAI Announce Strategic Partnership to Deploy 6 Gigawatts of AMD GPUs","publisher":"AMD","type":"press-release","date":"2025-10-06"}}
{"id":"2025-10-06-openai-agentkit","date":"2025-10-06","precision":"day","title":"OpenAI ships AgentKit, a toolkit for building AI agents","summary":"Agent Builder, ChatKit, Evals for Agents and a Connector Registry bundle into building blocks for taking an agent from prototype to production. Announced at DevDay alongside the Apps SDK.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://techcrunch.com/2025/10/06/openai-launches-agentkit-to-help-developers-build-and-ship-ai-agents/","title":"OpenAI launches AgentKit to help developers build and ship AI agents","publisher":"TechCrunch","type":"news","date":"2025-10-06"}}
{"id":"2025-10-06-openai-apps-sdk","date":"2025-10-06","precision":"day","title":"OpenAI opens ChatGPT to third-party apps at DevDay","summary":"Developers can now build interactive apps that run inside ChatGPT conversations, previewed at DevDay with Booking.com, Expedia, Spotify, Figma, Coursera, Zillow and Canva. A preview of the Apps SDK lets anyone build one, replacing the standalone GPT Store with apps embedded directly in chat.","category":"product","orgs":["OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2025/10/06/openai-launches-apps-inside-of-chatgpt/","title":"OpenAI launches apps inside of ChatGPT","publisher":"TechCrunch","type":"news","date":"2025-10-06"}}
{"id":"2025-10-07-deloitte-refunds-australia","date":"2025-10-07","precision":"day","title":"Deloitte refunds a government for a report the model made up","summary":"A $290,000 review of Australia's welfare compliance system cited academic papers that do not exist and a quote no federal court judge ever wrote. A Sydney University researcher noticed, Deloitte reissued the report disclosing it had used Azure OpenAI, and handed back the last instalment.","category":"culture","orgs":["Deloitte"],"tier":3,"source":{"url":"https://fortune.com/2025/10/07/deloitte-ai-australia-government-report-hallucinations-technology-290000-refund/","title":"Deloitte was caught using AI in $290,000 report to help the Australian government crack down on welfare after a researcher flagged hallucinations","publisher":"Fortune","author":"Nino Paoli","type":"news","date":"2025-10-07"}}
{"id":"2025-10-07-gemini-2-5-computer-use","date":"2025-10-07","precision":"day","title":"Google ships a model that drives a browser","summary":"Gemini 2.5 Computer Use clicks, types and scrolls its way through web interfaces, built on Gemini 2.5 Pro and aimed at browsers first with mobile still emerging. Google claims the lead on Online-Mind2Web, WebVoyager and AndroidWorld.","category":"model","orgs":["Google"],"tier":3,"source":{"url":"https://blog.google/technology/google-deepmind/gemini-computer-use-model/","title":"Introducing the Gemini 2.5 Computer Use model","publisher":"Google","type":"announcement","date":"2025-10-07"}}
{"id":"2025-10-07-nexperia-court-suspension","date":"2025-10-07","precision":"day","title":"A Dutch court suspends Nexperia's CEO and freezes Wingtech's control","summary":"The Enterprise Chamber found valid reasons to doubt sound management under Zhang Xuezheng, suspended him and placed Wingtech's voting shares in Nexperia under an independent administrator. China retaliated within days by blocking Nexperia's Chinese plant from exporting finished chips, disrupting automakers worldwide.","category":"policy","orgs":["Nexperia"],"tier":2,"source":{"url":"https://www.nexperia.com/about/news-events/press-releases/update-on-company-developments","title":"Update on company developments","publisher":"Nexperia","type":"press-release","date":"2025-10-14"},"references":[{"url":"https://www.government.nl/latest/news/2025/10/12/minister-of-economic-affairs-invokes-goods-availability-act","title":"Minister of Economic Affairs invokes Goods Availability Act","publisher":"Government of the Netherlands","type":"press-release","date":"2025-10-12"}]}
{"id":"2025-10-07-qualcomm-acquires-arduino","date":"2025-10-07","precision":"day","title":"Qualcomm agrees to acquire Arduino","summary":"Arduino agreed to join Qualcomm as an independent subsidiary, pending regulatory approval, ending two decades as an independent open hardware company. The companies unveiled UNO Q, a board pairing a Linux capable Qualcomm processor with a microcontroller, aimed at Arduino's 33 million person community.","category":"acquisition","orgs":["Qualcomm","Arduino"],"tier":3,"source":{"url":"https://blog.arduino.cc/2025/10/07/a-new-chapter-for-arduino-with-qualcomm-uno-q-and-you/","title":"A new chapter for Arduino – with Qualcomm, UNO Q, and you!","publisher":"Arduino","type":"announcement","date":"2025-10-07"},"references":[{"url":"https://arstechnica.com/gadgets/2025/10/arduino-retains-its-brand-and-mission-following-acquisition-by-qualcomm/","title":"Arduino retains its brand and mission following acquisition by Qualcomm","publisher":"Ars Technica","type":"news","date":"2025-10-07"}]}
{"id":"2025-10-09-data-poisoning-scale-invariance","date":"2025-10-09","precision":"day","title":"A few hundred documents can poison a model of any size","summary":"Anthropic, the UK AI Security Institute and the Alan Turing Institute found that the number of malicious training documents needed to install a backdoor stays roughly constant as models scale. The intuition that bigger models are harder to poison turned out to be backwards.","category":"research","orgs":["Anthropic","UK AI Security Institute"],"tier":3,"source":{"url":"https://www.anthropic.com/research/small-samples-poison","title":"A small number of samples can poison LLMs of any size","publisher":"Anthropic","type":"paper","date":"2025-10-09"}}
{"id":"2025-10-09-reflection-ai-series-b","date":"2025-10-09","precision":"day","title":"Reflection AI raises $2 billion to build an open frontier lab","summary":"The round values the year old startup at $8 billion, fifteen times its valuation seven months earlier. Founded by former Google DeepMind researchers, Reflection AI plans to publish model weights while keeping training data and pipelines proprietary, pitched as a Western answer to DeepSeek.","category":"funding","orgs":["Reflection AI"],"tier":2,"source":{"url":"https://techcrunch.com/2025/10/09/reflection-raises-2b-to-be-americas-open-frontier-ai-lab-challenging-deepseek/","title":"Reflection raises $2B to be America's open frontier AI lab, challenging DeepSeek","publisher":"TechCrunch","type":"news","date":"2025-10-09"}}
{"id":"2025-10-13-karpathy-nanochat","date":"2025-10-13","precision":"day","title":"Karpathy ships nanochat, a ChatGPT clone trainable for about $100","summary":"Andrej Karpathy released a complete, minimal codebase for training and serving a ChatGPT style model on rented GPUs for roughly $100, in the spirit of the from-scratch AI course his company Eureka Labs is building. The resulting 561 million parameter model reasons about as well as GPT-2.","category":"open-source","orgs":["Eureka Labs"],"tier":3,"source":{"url":"https://github.com/karpathy/nanochat/discussions/1","title":"Introducing nanochat: The best ChatGPT that $100 can buy.","publisher":"Andrej Karpathy","type":"announcement","date":"2025-10-13","author":"Andrej Karpathy"}}
{"id":"2025-10-13-openai-broadcom-10-gigawatts","date":"2025-10-13","precision":"day","title":"OpenAI and Broadcom agree to co-develop 10 gigawatts of AI chips","summary":"OpenAI committed to a multiyear purchase of custom AI accelerators and networking gear from Broadcom, with hardware rolling out through the end of 2029. It follows OpenAI's 6 gigawatt AMD deal a week earlier, adding to tens of gigawatts of announced compute commitments.","category":"deal","orgs":["OpenAI","Broadcom"],"tier":2,"source":{"url":"https://fortune.com/2025/10/13/openai-broadcom-chipmaker-artificial-intelligence-data-center/","title":"OpenAI-Broadcom agreement sends shares of chipmaker soaring","publisher":"Fortune","type":"news","date":"2025-10-13"}}
{"id":"2025-10-15-apple-m5-chip","date":"2025-10-15","precision":"day","title":"Apple ships the M5 chip","summary":"The M5 brings a GPU with a neural accelerator in each core, more than quadrupling peak GPU compute for AI workloads over the M4. It ships first in the 14 inch MacBook Pro, iPad Pro and Vision Pro.","category":"hardware","orgs":["Apple"],"tier":3,"source":{"url":"https://www.apple.com/newsroom/2025/10/apple-unleashes-m5-the-next-big-leap-in-ai-performance-for-apple-silicon/","title":"Apple unleashes M5, the next big leap in AI performance for Apple silicon","publisher":"Apple","type":"press-release","date":"2025-10-15"}}
{"id":"2025-10-15-claude-haiku-4-5","date":"2025-10-15","precision":"day","title":"Claude Haiku 4.5 puts five-month-old frontier coding at a third of the price","summary":"Anthropic's small model matches Sonnet 4 on coding, reaches 90 percent of Sonnet 4.5 on agentic coding, and runs four to five times faster than Sonnet 4.5 at $1 and $5 per million tokens. The frontier of May was the cheap tier by October.","category":"model","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/claude-haiku-4-5","title":"Introducing Claude Haiku 4.5","publisher":"Anthropic","type":"announcement","date":"2025-10-15"}}
{"id":"2025-10-16-agent-skills","date":"2025-10-16","precision":"day","title":"Anthropic ships Skills","summary":"Folders of instructions, scripts and resources that Claude loads only when a task calls for them, in one format that works across the apps, the API and Claude Code. Composable, portable between surfaces, and paged in only when relevant.","category":"product","orgs":["Anthropic"],"tier":2,"source":{"url":"https://claude.com/blog/skills","title":"Introducing Agent Skills","publisher":"Anthropic","type":"announcement","date":"2025-10-16"}}
{"id":"2025-10-16-doordash-waymo-autonomous-delivery","date":"2025-10-16","precision":"day","title":"DoorDash and Waymo begin testing autonomous deliveries in Phoenix","summary":"DoorDash customers in Metro Phoenix can opt into having DashMart orders carried by a driverless Waymo vehicle instead of a human Dasher, with a commercial rollout planned for later in 2025.","category":"product","orgs":["DoorDash","Waymo"],"tier":3,"source":{"url":"https://ir.doordash.com/news/news-details/2025/DoorDash-and-Waymo-Launch-Autonomous-Delivery-Service-and-Introduce-Limited-Time-Waymo-Promotion-for-DashPass-Members/default.aspx","title":"DoorDash and Waymo Launch Autonomous Delivery Service and Introduce Limited-Time Waymo Promotion for DashPass Members","publisher":"DoorDash","type":"press-release","date":"2025-10-16"}}
{"id":"2025-10-20-claude-code-on-the-web","date":"2025-10-20","precision":"day","title":"Anthropic ships Claude Code on the web","summary":"A research preview lets Pro and Max subscribers assign coding tasks to Claude that run in isolated cloud sandboxes instead of a local terminal, so several tasks can run in parallel across repositories from a browser or the iOS app.","category":"product","orgs":["Anthropic"],"tier":3,"source":{"url":"https://claude.com/blog/claude-code-on-the-web","title":"Claude Code on the web","publisher":"Anthropic","type":"announcement","date":"2025-10-20"}}
{"id":"2025-10-20-deepseek-ocr","date":"2025-10-20","precision":"day","title":"DeepSeek compresses text into pictures of text","summary":"DeepSeek-OCR treats a vision encoder as a compression scheme for context, packing a page into between 64 and 400 vision tokens depending on resolution. Released MIT licensed, and read less as an OCR model than as an argument about what context windows could cost.","category":"model","secondaryCategories":["open-source"],"orgs":["DeepSeek"],"tier":3,"source":{"url":"https://github.com/deepseek-ai/DeepSeek-OCR","title":"deepseek-ai/DeepSeek-OCR","publisher":"DeepSeek","type":"docs","retrieved":"2026-08-20","locator":"[2025/10/20] We release DeepSeek-OCR, a model to investigate the role of vision encoders from an LLM-centric viewpoint."}}
{"id":"2025-10-21-chatgpt-atlas","date":"2025-10-21","precision":"day","title":"OpenAI ships a browser built around ChatGPT","summary":"Atlas launched on macOS with ChatGPT in the sidebar, memory of the pages you visited, and an agent mode for paying tiers that works through a site on your behalf. Three months after Comet, the browser had become the thing the labs fight over.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://openai.com/index/introducing-chatgpt-atlas/","title":"Introducing ChatGPT Atlas","publisher":"OpenAI","type":"announcement","date":"2025-10-21"},"references":[{"url":"https://techcrunch.com/2025/10/21/openai-launches-an-ai-powered-browser-chatgpt-atlas/","title":"OpenAI launches an AI-powered browser: ChatGPT Atlas","publisher":"TechCrunch","type":"news","date":"2025-10-21"},{"url":"https://simonwillison.net/2025/Oct/21/introducing-chatgpt-atlas/","title":"Introducing ChatGPT Atlas","publisher":"Simon Willison","author":"Simon Willison","type":"analysis","date":"2025-10-21"}]}
{"id":"2025-10-22-ai-assistants-misrepresent-news","date":"2025-10-22","precision":"day","title":"AI assistants get the news wrong 45 percent of the time","summary":"The European Broadcasting Union and the BBC put ChatGPT, Gemini, Copilot and Perplexity in front of news questions in 14 languages across 22 public broadcasters, and 45 percent of answers carried at least one significant problem. Sourcing was the worst of it at 31 percent, and Gemini's sourcing failed in 76 percent of its answers.","category":"research","orgs":["European Broadcasting Union"],"tier":2,"source":{"url":"https://www.aljazeera.com/news/2025/10/22/ai-models-misrepresent-news-events-nearly-half-the-time-study-says","title":"AI models misrepresent news events nearly half the time, study says","publisher":"Al Jazeera","author":"John Power","type":"news","date":"2025-10-22"},"references":[{"url":"https://www.ebu.ch/news/2025/10/ai-s-systemic-distortion-of-news-is-consistent-across-languages-and-territories-international-study-by-public-service-broadcaste","title":"Largest study of its kind shows AI assistants misrepresent news content 45% of the time – regardless of language or territory","publisher":"European Broadcasting Union","type":"report","retrieved":"2026-08-20"}]}
{"id":"2025-10-22-meta-ai-superintelligence-layoffs","date":"2025-10-22","precision":"day","title":"Meta cuts 600 jobs from its AI Superintelligence Labs","summary":"The cuts came months after Meta paid large packages to recruit more than 50 researchers into the unit. Chief AI officer Alexandr Wang said a smaller team would need fewer conversations to reach a decision; most affected employees were expected to find other roles inside the company.","category":"company","orgs":["Meta"],"tier":3,"source":{"url":"https://techcrunch.com/2025/10/22/meta-cuts-600-ai-jobs-amid-ongoing-reorganization/","title":"Meta cuts 600 AI jobs amid ongoing reorganization","publisher":"TechCrunch","type":"news","date":"2025-10-22"}}
{"id":"2025-10-23-openai-acquires-sky-app","date":"2025-10-23","precision":"day","title":"OpenAI acquires Sky, a natural language interface for Mac","summary":"OpenAI bought Software Applications, a startup whose unreleased Sky product floats over the desktop to see the screen and act in other Mac apps. Its founders previously built Workflow before Apple acquired it and turned it into Shortcuts.","category":"acquisition","orgs":["OpenAI","Software Applications"],"tier":3,"source":{"url":"https://techcrunch.com/2025/10/23/openai-buys-sky-an-ai-interface-for-mac/","title":"OpenAI buys Sky, an AI interface for Mac","publisher":"TechCrunch","type":"news","date":"2025-10-23"}}
{"id":"2025-10-24-crusoe-series-e","date":"2025-10-24","precision":"day","title":"Crusoe raises $1.375 billion to build AI data centers","summary":"The Series E values the infrastructure company above $10 billion, co-led by Valor Equity Partners and Mubadala Capital with Nvidia among the backers. Crusoe pairs dedicated power generation with gigawatt scale campuses, including projects underway in Texas and Wyoming.","category":"funding","orgs":["Crusoe"],"tier":3,"source":{"url":"https://www.crusoe.ai/resources/newsroom/crusoe-announces-series-e-funding","title":"Crusoe, the AI factory company, raising $1.375 billion at a valuation above $10 billion to power the future of AI infrastructure","publisher":"Crusoe","type":"press-release","date":"2025-10-24"}}
{"id":"2025-10-27-claude-for-excel","date":"2025-10-27","precision":"day","title":"Anthropic ships Claude for Excel","summary":"A sidebar lets Claude read, audit and edit live Excel workbooks while preserving formulas and structure, released in beta to about 1,000 Max, Team and Enterprise users alongside a financial data partnership with LSEG.","category":"product","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/advancing-claude-for-financial-services","title":"Advancing Claude for Financial Services","publisher":"Anthropic","type":"announcement","date":"2025-10-27"},"references":[{"url":"https://venturebeat.com/technology/anthropic-rolls-out-claude-ai-for-finance-integrates-with-excel-to-rival","title":"Anthropic rolls out Claude AI for finance, integrates with Excel to rival Microsoft Copilot","publisher":"VentureBeat","type":"news","date":"2025-10-27"}]}
{"id":"2025-10-27-grokipedia-launches","date":"2025-10-27","precision":"day","title":"xAI launches Grokipedia, an AI generated Wikipedia rival","summary":"Elon Musk announced Grokipedia version 0.1 on X, an AI generated encyclopedia built by xAI with about 885,000 articles at launch, roughly 12 percent of English Wikipedia's count. Independent reviewers quickly found plagiarized passages and skewed entries on political topics.","category":"product","orgs":["xAI"],"tier":3,"source":{"url":"https://www.theregister.com/software/2025/10/28/grokipedia_is_such_a_mess_even_grok_thinks_its_untrustworthy/1472249","title":"Elon Musk's Grokipedia launches, filled to the brim with plagiarism and AI slop","publisher":"The Register","type":"news","date":"2025-10-28"}}
{"id":"2025-10-27-mercor-series-c","date":"2025-10-27","precision":"day","title":"Mercor raises $350 million to supply human data for AI labs","summary":"The Series C values the labor marketplace at $10 billion, five times its prior round, led by Felicis with Benchmark, General Catalyst and Robinhood Ventures joining. Mercor connects expert professionals with AI labs that pay them to train frontier models.","category":"funding","orgs":["Mercor"],"tier":3,"source":{"url":"https://mercor.com/blog/series-c/","title":"Unlocking human potential in the AI economy","publisher":"Mercor","type":"announcement","date":"2025-10-27"}}
{"id":"2025-10-28-nvidia-invests-in-nokia","date":"2025-10-28","precision":"day","title":"Nvidia invests $1 billion in Nokia for AI-RAN hardware","summary":"NVIDIA agreed to buy $1 billion of Nokia stock at $6.01 a share, becoming the telecom equipment maker's second largest shareholder, as the two build AI-RAN radio hardware for 6G networks with T-Mobile field trials planned for 2026.","category":"deal","orgs":["NVIDIA","Nokia"],"tier":3,"source":{"url":"https://investor.nvidia.com/news/press-release-details/2025/NVIDIA-and-Nokia-to-Pioneer-the-AI-Platform-for-6G--Powering-Americas-Return-to-Telecommunications-Leadership/default.aspx","title":"NVIDIA and Nokia to Pioneer the AI Platform for 6G — Powering America's Return to Telecommunications Leadership","publisher":"NVIDIA","type":"press-release","date":"2025-10-28"}}
{"id":"2025-10-28-openai-public-benefit-corporation","date":"2025-10-28","precision":"day","title":"OpenAI's for-profit becomes a public benefit corporation","summary":"The recapitalization left the nonprofit foundation holding about 26 percent of OpenAI Group PBC and Microsoft about 27 percent, after a year of negotiation with two state attorneys general. The structure that produced the 2023 board crisis was finally rewritten.","category":"company","orgs":["OpenAI","Microsoft"],"tier":1,"source":{"url":"https://openai.com/index/built-to-benefit-everyone/","title":"Built to benefit everyone","publisher":"OpenAI","type":"announcement","date":"2025-10-28"},"references":[{"url":"https://news.delaware.gov/2025/10/28/ag-jennings-completes-review-of-openai-recapitalization/","title":"AG Jennings completes review of OpenAI recapitalization","publisher":"State of Delaware","type":"press-release","date":"2025-10-28"},{"url":"https://www.theguardian.com/technology/2025/oct/28/openai-for-profit-restructuring","title":"OpenAI completes conversion to for-profit business after lengthy negotiations","publisher":"The Guardian","type":"news","date":"2025-10-28"}]}
{"id":"2025-10-29-character-ai-bars-minors","date":"2025-10-29","precision":"day","title":"Character.AI bars open ended chat for users under 18","summary":"Character.AI said users under 18 will lose open ended chat with its characters by November 25, with daily time capped in the interim, and is standing up an independent AI safety lab and age verification. The change follows lawsuits over chatbot conversations tied to teen self harm.","category":"culture","orgs":["Character.AI"],"tier":3,"source":{"url":"https://blog.character.ai/u18-chat-announcement/","title":"Taking Bold Steps to Keep Teen Users Safe on Character.AI","publisher":"Character.AI","type":"announcement","date":"2025-10-29"},"references":[{"url":"https://apnews.com/article/characterai-kids-minors-18-ban-chatbot-5d203e9f22c62c153936ccc776a0ed09","title":"Character.AI is banning minors from interacting with its chatbots","publisher":"AP News","type":"news","date":"2025-10-29"}]}
{"id":"2025-10-29-cursor-2-0","date":"2025-10-29","precision":"day","title":"Cursor 2.0 ships Composer, a coding model built for speed","summary":"Anysphere's Cursor 2.0 redesigns the editor around running several coding agents in parallel across git worktrees, and ships Composer, a model the company says completes most tasks in under 30 seconds at roughly four times the speed of comparably capable models.","category":"product","orgs":["Anysphere"],"tier":3,"source":{"url":"https://cursor.com/blog/2-0","title":"Introducing Cursor 2.0 and Composer","publisher":"Cursor","type":"announcement","date":"2025-10-29"}}
{"id":"2025-10-29-nvidia-five-trillion","date":"2025-10-29","precision":"day","title":"Nvidia becomes the first $5 trillion company","summary":"Nine months after losing $600 billion in a day over the theory that efficient models need fewer chips, Nvidia closed above five trillion dollars. The market had settled on cheaper inference meaning more of it, not less.","category":"hardware","orgs":["NVIDIA"],"tier":2,"source":{"url":"https://www.cnbc.com/2025/10/29/nvidia-on-track-to-hit-historic-5-trillion-valuation-amid-ai-rally.html","title":"Nvidia becomes first company to reach $5 trillion valuation","publisher":"CNBC","type":"news","date":"2025-10-29"},"references":[{"url":"https://www.bbc.com/news/articles/cp8e970vn5vo","title":"Nvidia hits new milestone as world's first $5tn company","publisher":"BBC News","type":"news","date":"2025-10-29"}]}
{"id":"2025-10-29-umg-udio-settlement","date":"2025-10-29","precision":"day","title":"Universal Music settles its Udio lawsuit and signs a licensing deal","summary":"Universal Music Group and Udio resolved the copyright suit UMG filed in 2024 over unlicensed training data, agreeing to launch a new licensed AI music platform where artists control how their voice and style can be used. Udio pulled its existing downloads during the transition.","category":"deal","orgs":["Universal Music Group","Udio"],"tier":3,"source":{"url":"https://udio.com/blog/a-new-era","title":"A New Era of Music","publisher":"Udio","type":"announcement","date":"2025-10-29"}}
{"id":"2025-10-30-kimi-linear","date":"2025-10-30","precision":"day","title":"Moonshot AI publishes the Kimi Linear attention architecture","summary":"Kimi Linear replaces most attention layers with a hybrid linear mechanism called Kimi Delta Attention, cutting KV cache use by up to 75 percent and lifting decoding throughput up to 6 times at a million token context while matching full attention on benchmarks.","category":"research","orgs":["Moonshot AI"],"tier":3,"source":{"url":"https://arxiv.org/abs/2510.26692","title":"Kimi Linear: An Expressive, Efficient Attention Architecture","publisher":"arXiv","type":"paper","date":"2025-10-30"}}
{"id":"2025-11-03-openai-aws-deal","date":"2025-11-03","precision":"day","title":"OpenAI turns to AWS for $38 billion in compute","summary":"A seven-year, $38 billion deal gives OpenAI immediate access to hundreds of thousands of Nvidia GPUs on AWS, with all planned capacity live by the end of 2026. It was OpenAI's clearest move yet outside Microsoft's Azure, which had been its exclusive cloud until earlier in the year.","category":"deal","orgs":["OpenAI","Amazon"],"tier":2,"source":{"url":"https://www.aljazeera.com/economy/2025/11/3/openai-amazon-sign-38bn-ai-deal","title":"OpenAI, Amazon sign $38bn AI deal","publisher":"Al Jazeera","type":"news","date":"2025-11-03"},"references":[{"url":"https://www.aboutamazon.com/news/aws/aws-open-ai-workloads-compute-infrastructure","title":"AWS and OpenAI announce multi-year strategic partnership","publisher":"Amazon","type":"press-release"}]}
{"id":"2025-11-03-project-rainier-online","date":"2025-11-03","precision":"day","title":"AWS activates Project Rainier, a 500,000-chip AI supercomputer","summary":"AWS switched on Project Rainier, a cluster of nearly 500,000 custom Trainium2 chips, with Anthropic already training and serving Claude on it. AWS said the cluster would grow past a million chips by the end of the year.","category":"hardware","orgs":["Amazon","Anthropic"],"tier":3,"source":{"url":"https://aws.amazon.com/blogs/aws/aws-weekly-roundup-project-rainier-online-amazon-nova-amazon-bedrock-and-more-november-3-2025","title":"AWS Weekly Roundup: Project Rainier online, Amazon Nova, Amazon Bedrock, and more","publisher":"AWS","type":"announcement","date":"2025-11-03"}}
{"id":"2025-11-06-burry-shorts-nvidia-palantir","date":"2025-11-06","precision":"day","title":"Michael Burry discloses billion-dollar bets against Nvidia and Palantir","summary":"Scion Asset Management's quarterly filing showed put options with more than a billion dollars of notional value against Nvidia and Palantir, the highest-profile of the year's public bets that AI stocks had run ahead of themselves.","category":"culture","orgs":["NVIDIA","Palantir"],"tier":3,"source":{"url":"https://www.fool.com/investing/2025/11/06/michael-burry-bet-against-ai-stock-palantir-nvidia","title":"Billionaire Michael Burry Bets Big Against AI Stocks Palantir and Nvidia","publisher":"The Motley Fool","type":"news","date":"2025-11-06"},"references":[{"url":"https://sherwood.news/markets/michael-burry-big-short-discloses-1-1-billion-options-bet-against-nvidia-palantir-puts/","title":"Michael Burry, of “Big Short” fame, discloses options bets against Nvidia and Palantir","publisher":"Sherwood News","type":"news","date":"2025-11-13"}]}
{"id":"2025-11-06-ironwood-tpu-ga","date":"2025-11-06","precision":"day","title":"Google unveils Ironwood, its seventh-generation TPU","summary":"Google detailed Ironwood, a seventh-generation TPU with six times the memory of its predecessor, built for the inference-heavy workloads AI companies increasingly run. Anthropic said it plans to use up to a million of them.","category":"hardware","orgs":["Google"],"tier":3,"source":{"url":"https://cloud.google.com/blog/products/compute/ironwood-tpus-and-new-axion-based-vms-for-your-ai-workloads/","title":"Ironwood TPUs and new Axion-based VMs, generally available soon, to power your most demanding AI workloads","publisher":"Google Cloud","type":"announcement","date":"2025-11-06"}}
{"id":"2025-11-06-kimi-k2-thinking","date":"2025-11-06","precision":"day","title":"Kimi K2 Thinking tops the open-weight field","summary":"A trillion-parameter reasoning model from Moonshot AI, released with open weights and beating several closed frontier models on agentic benchmarks. The US government's AI standards center published its own evaluation five weeks later.","category":"model","secondaryCategories":["open-source"],"orgs":["Moonshot AI"],"tier":2,"source":{"url":"https://moonshotai.github.io/Kimi-K2/thinking.html","title":"Kimi K2 Thinking","publisher":"Moonshot AI","type":"announcement","date":"2025-11-06"},"references":[{"url":"https://www.nist.gov/news-events/news/2025/12/caisi-evaluation-kimi-k2-thinking","title":"CAISI Evaluation of Kimi K2 Thinking","publisher":"NIST","type":"report","date":"2025-12-12"},{"url":"https://www.cnbc.com/2025/11/06/alibaba-backed-moonshot-releases-new-ai-model-kimi-k2-thinking.html","title":"Alibaba-backed Moonshot releases new AI model Kimi K2 Thinking","publisher":"CNBC","type":"news","date":"2025-11-06"}]}
{"id":"2025-11-11-softbank-sells-nvidia","date":"2025-11-11","precision":"day","title":"SoftBank sells all of its Nvidia to pay for OpenAI","summary":"All 32.1 million shares went in October for $5.8 billion, and SoftBank's finance chief gave the reason outright: more than $30 billion had to go into OpenAI this year, so existing assets had to be sold to find it. The projected total by year end was $34.7 billion.","category":"funding","orgs":["SoftBank","NVIDIA","OpenAI"],"tier":3,"source":{"url":"https://fortune.com/2025/11/11/softbank-nvidia-openai-masayoshi-son-sam-altman-investment/","title":"Softbank dumps its entire Nvidia portfolio worth $5.8 billion as its CEO goes all-in on OpenAI to the tune of $30 billion","publisher":"Fortune","author":"Dave Smith","type":"news","date":"2025-11-11"}}
{"id":"2025-11-12-anthropic-infrastructure-investment","date":"2025-11-12","precision":"day","title":"Anthropic commits $50 billion to US data centers","summary":"Anthropic will spend $50 billion building AI data centers with Fluidstack in Texas and New York, its first move into owning infrastructure rather than only renting compute from cloud partners.","category":"deal","orgs":["Anthropic","Fluidstack"],"tier":3,"source":{"url":"https://www.anthropic.com/news/anthropic-invests-50-billion-in-american-ai-infrastructure","title":"Anthropic invests $50 billion in American AI infrastructure","publisher":"Anthropic","type":"announcement","date":"2025-11-12"},"references":[{"url":"https://siliconangle.com/2025/11/12/anthropic-spend-50b-us-data-center-infrastructure/","title":"Anthropic to spend $50B on US data center infrastructure","publisher":"SiliconANGLE","type":"news","date":"2025-11-12"}]}
{"id":"2025-11-12-gpt-5-1","date":"2025-11-12","precision":"day","title":"OpenAI ships GPT-5.1 with a warmer default","summary":"Instant and Thinking variants, reasoning that decides for itself how long to think, and eight selectable personalities. The headline feature of a frontier release was now its tone of voice.","category":"model","orgs":["OpenAI"],"tier":3,"source":{"url":"https://openai.com/index/gpt-5-1/","title":"GPT-5.1: A smarter, more conversational ChatGPT","publisher":"OpenAI","type":"announcement","date":"2025-11-12","archiveUrl":"https://web.archive.org/web/20251112194902/https://openai.com/index/gpt-5-1/"},"references":[{"url":"https://www.macrumors.com/2025/11/12/openai-chatgpt-5-1-launch/","title":"OpenAI Launches Smarter, More Conversational ChatGPT 5.1","publisher":"MacRumors","type":"news","date":"2025-11-12"}]}
{"id":"2025-11-13-anthropic-ai-espionage","date":"2025-11-13","precision":"day","title":"Anthropic reports the first AI-run espionage campaign","summary":"A Chinese state-sponsored group turned Claude Code on roughly thirty targets and succeeded against a few, with Anthropic putting the model's share of the work at 80 to 90 percent and human involvement at four to six decision points. Security researchers objected that the report shipped without a single indicator of compromise anyone could act on.","category":"research","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/news/disrupting-AI-espionage","title":"Disrupting the first reported AI-orchestrated cyber espionage campaign","publisher":"Anthropic","type":"report","date":"2025-11-13"},"references":[{"url":"https://djnn.sh/posts/anthropic-s-paper-smells-like-bullshit/","title":"anthropic's paper smells like bullshit","publisher":"djnn","author":"djnn","type":"analysis","date":"2025-11-16"}]}
{"id":"2025-11-13-cursor-series-d","date":"2025-11-13","precision":"day","title":"Cursor raises $2.3 billion at a $29.3 billion valuation","summary":"Anysphere's Series D valued the AI coding editor Cursor at $29.3 billion, with Nvidia and Google joining existing investors Accel, Thrive and Andreessen Horowitz. The company said it had crossed a billion dollars in annualized revenue with a team of just over 300.","category":"funding","orgs":["Anysphere"],"tier":3,"source":{"url":"https://cursor.com/blog/series-d","title":"Past, Present, and Future","publisher":"Cursor","type":"announcement","date":"2025-11-13"}}
{"id":"2025-11-17-grok-4-1","date":"2025-11-17","precision":"day","title":"xAI ships Grok 4.1, tuned for warmth over benchmarks","summary":"Grok 4.1 followed a two-week silent rollout against live traffic before its public launch, with xAI emphasizing emotional read and fewer hallucinations over the usual benchmark chase.","category":"model","orgs":["xAI"],"tier":3,"source":{"url":"https://x.ai/news/grok-4-1","title":"Grok 4.1","publisher":"xAI","type":"announcement","date":"2025-11-17"}}
{"id":"2025-11-18-gemini-3","date":"2025-11-18","precision":"day","title":"Gemini 3 ships into Search on day one","summary":"Google released Gemini 3 Pro in preview and put it into AI Mode in Search, the Gemini app, developer tools and Vertex AI on launch day. Its published scores covered different tool conditions and evaluation harnesses, so they describe specific tests rather than a universal ranking.","category":"model","orgs":["Google DeepMind"],"tier":2,"source":{"url":"https://blog.google/products/gemini/gemini-3/","title":"Gemini 3: Introducing the latest Gemini AI model from Google","publisher":"Google","type":"announcement","date":"2025-11-18"}}
{"id":"2025-11-18-google-antigravity","date":"2025-11-18","precision":"day","title":"Google launches Antigravity, an agent-first coding platform","summary":"Built on the Windsurf team Google acquired in July, Antigravity lets agents plan, run and verify multi-step coding tasks across the editor, terminal and browser. It supports Gemini 3 alongside Anthropic's and OpenAI's models.","category":"product","orgs":["Google"],"tier":3,"source":{"url":"https://venturebeat.com/ai/google-antigravity-introduces-agent-first-architecture-for-asynchronous","title":"Google Antigravity introduces agent-first architecture for asynchronous, verifiable coding workflows","publisher":"VentureBeat","type":"news","date":"2025-11-18"},"references":[{"url":"https://developers.googleblog.com/build-with-google-antigravity-our-new-agentic-development-platform/","title":"Build with Google Antigravity, our new agentic development platform","publisher":"Google Developers Blog","type":"docs","date":"2025-11-20"}]}
{"id":"2025-11-18-microsoft-nvidia-invest-in-anthropic","date":"2025-11-18","precision":"day","title":"Microsoft and Nvidia put up to $15 billion into Anthropic","summary":"Nvidia up to $10 billion, Microsoft up to $5 billion, and Anthropic committing $30 billion straight back to Azure. Claude became the only frontier model sold on all three major clouds.","category":"funding","orgs":["Microsoft","NVIDIA","Anthropic"],"tier":2,"source":{"url":"https://blogs.microsoft.com/blog/2025/11/18/microsoft-nvidia-and-anthropic-announce-strategic-partnerships/","title":"Microsoft, NVIDIA and Anthropic announce strategic partnerships","publisher":"Microsoft","type":"announcement","date":"2025-11-18"}}
{"id":"2025-11-19-gpt-5-1-codex-max","date":"2025-11-19","precision":"day","title":"OpenAI ships GPT-5.1-Codex-Max for long coding sessions","summary":"The model adds compaction, discarding stale context so it can keep working across sessions that would otherwise exceed its context window, and OpenAI said it completed a 24-hour task internally. It edged out Gemini 3 Pro on SWE-bench Verified.","category":"model","orgs":["OpenAI"],"tier":3,"source":{"url":"https://venturebeat.com/technology/openai-debuts-gpt-5-1-codex-max-coding-model-and-it-already-completed-a-24","title":"OpenAI debuts GPT-5.1-Codex-Max coding model and it already completed a 24-hour task internally","publisher":"VentureBeat","type":"news","date":"2025-11-19"}}
{"id":"2025-11-19-larry-summers-resigns-openai-board","date":"2025-11-19","precision":"day","title":"Larry Summers resigns from OpenAI's board over Epstein emails","summary":"Congress released emails showing years of contact between Summers and Jeffrey Epstein, and Summers resigned from OpenAI's board the same day. He also stepped back from Harvard, the New York Times and several other posts.","category":"company","orgs":["OpenAI"],"tier":3,"source":{"url":"https://www.cbsnews.com/news/larry-summers-resigns-openai-board-jeffrey-epstein-emails/","title":"Larry Summers resigns from OpenAI board and other roles after Epstein emails released","publisher":"CBS News","type":"news","date":"2025-11-19"}}
{"id":"2025-11-19-lecun-leaves-meta","date":"2025-11-19","precision":"day","title":"Yann LeCun leaves Meta to bet against language models","summary":"Meta's chief AI scientist departed after twelve years to start a company on world models, having spent much of that time arguing publicly that scaling language models would not reach human-level intelligence. Meta had just spent $14.3 billion building a lab around the opposite thesis.","category":"company","orgs":["Meta"],"tier":3,"source":{"url":"https://www.cnbc.com/2025/11/19/meta-chief-ai-scientist-yann-lecun-is-leaving-the-company-.html","title":"Meta chief AI scientist Yann LeCun is leaving the company","publisher":"CNBC","type":"news","date":"2025-11-19"}}
{"id":"2025-11-19-sam-3","date":"2025-11-19","precision":"day","title":"Segment Anything learns to take a description","summary":"SAM 3 finds and tracks every instance matching a phrase like red baseball cap, where SAM 1 and SAM 2 returned one object per prompt. Meta released the weights, an open-vocabulary benchmark and a paper, along with SAM 3D for rebuilding objects and human bodies from a single image.","category":"model","orgs":["Meta"],"tier":3,"source":{"url":"https://about.fb.com/news/2025/11/new-sam-models-detect-objects-create-3d-reconstructions/","title":"New Segment Anything Models Make it Easier to Detect Objects and Create 3D Reconstructions","publisher":"Meta","type":"announcement","date":"2025-11-19"}}
{"id":"2025-11-19-workday-acquires-pipedream","date":"2025-11-19","precision":"day","title":"Workday agrees to acquire the AI agent platform Pipedream","summary":"The deal gives Workday's AI agents access to Pipedream's more than 3,000 prebuilt connectors to outside apps like Slack, Jira and HubSpot, its third agent-infrastructure acquisition of the year after Sana and Flowise.","category":"acquisition","orgs":["Workday","Pipedream"],"tier":3,"source":{"url":"https://newsroom.workday.com/2025-11-19-Workday-Signs-Definitive-Agreement-to-Acquire-Pipedream","title":"Workday Signs Definitive Agreement to Acquire Pipedream","publisher":"Workday","type":"press-release","date":"2025-11-19"}}
{"id":"2025-11-20-nano-banana-pro","date":"2025-11-20","precision":"day","title":"Google puts Gemini 3 Pro behind an image model","summary":"Nano Banana Pro, formally Gemini 3 Pro Image, renders legible text in several languages, blends up to 14 images while holding five people consistent, and outputs at 4K. It landed in the Gemini app, Ads, Workspace, Vertex and NotebookLM on the same day.","category":"model","orgs":["Google"],"tier":3,"source":{"url":"https://blog.google/technology/ai/nano-banana-pro/","title":"Introducing Nano Banana Pro","publisher":"Google","type":"announcement","date":"2025-11-20"}}
{"id":"2025-11-24-claude-opus-4-5","date":"2025-11-24","precision":"day","title":"Anthropic releases Claude Opus 4.5 at a lower API price","summary":"Anthropic released Opus 4.5 at $5 per million input tokens, down from $15 for the prior Opus tier, and reported gains on coding evaluations.","category":"model","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/news/claude-opus-4-5","title":"Introducing Claude Opus 4.5","publisher":"Anthropic","type":"announcement","date":"2025-11-24"}}
{"id":"2025-11-24-clawdbot-repository-created","date":"2025-11-24","precision":"day","title":"The repository that becomes OpenClaw appears","summary":"Peter Steinberger created a project called Clawdbot: a personal agent that runs on hardware you own and takes instructions over a chat app. It went viral eight weeks later, was renamed twice inside a month after a trademark complaint, and now carries more than 386,000 stars.","category":"open-source","orgs":["OpenClaw"],"tier":3,"source":{"url":"https://github.com/openclaw/openclaw","title":"openclaw/openclaw","publisher":"OpenClaw","type":"docs","retrieved":"2026-08-20","locator":"Repository created 2025-11-24T10:16:47Z with 386,898 stars, per the GitHub API created_at and stargazers_count fields."}}
{"id":"2025-11-25-warner-music-suno-settlement","date":"2025-11-25","precision":"day","title":"Warner Music settles its Suno lawsuit and signs a licensing deal","summary":"Warner dropped its copyright suit against Suno in exchange for a licensing partnership, artist opt-in rights over AI use of their voice and likeness, and Suno's purchase of Warner's Songkick platform. Sony is now the only major label still suing both Suno and Udio.","category":"deal","orgs":["Warner Music Group","Suno"],"tier":3,"source":{"url":"https://techcrunch.com/2025/11/25/warner-music-signs-deal-with-ai-music-startup-suno-settles-lawsuit/","title":"Warner Music signs deal with AI music startup Suno, settles lawsuit","publisher":"TechCrunch","type":"news","date":"2025-11-25"}}
{"id":"2025-11-27-deepseek-math-v2","date":"2025-11-27","precision":"day","title":"DeepSeek's math model scores 118 of 120 on the Putnam","summary":"DeepSeekMath-V2 trains a verifier to catch its own reasoning errors, then trains a proof generator against that verifier as the reward signal. The 685-billion-parameter model, released under Apache 2.0, scored 118 of 120 on the 2024 Putnam and gold-level on IMO 2025.","category":"model","secondaryCategories":["open-source"],"orgs":["DeepSeek"],"tier":3,"source":{"url":"https://arxiv.org/abs/2511.22570","title":"DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning","publisher":"arXiv","type":"paper","date":"2025-11-27"},"references":[{"url":"https://simonwillison.net/2025/Nov/27/deepseek-math-v2/","title":"deepseek-ai/DeepSeek-Math-V2","publisher":"Simon Willison","type":"analysis","date":"2025-11-27"}]}
{"id":"2025-12-01-deepseek-v3-2","date":"2025-12-01","precision":"day","title":"DeepSeek puts gold-medal reasoning in MIT-licensed weights","summary":"Two models, V3.2 for agent work and V3.2-Speciale for pure reasoning, both open weight under MIT. DeepSeek claims Speciale took gold-level results at the IMO, the CMO, the ICPC World Finals and IOI 2025, and rivals Gemini 3 Pro.","category":"model","orgs":["DeepSeek"],"tier":2,"source":{"url":"https://api-docs.deepseek.com/news/news251201/","title":"DeepSeek-V3.2 Release","publisher":"DeepSeek","type":"announcement","date":"2025-12-01"},"references":[{"url":"https://huggingface.co/deepseek-ai/DeepSeek-V3.2","title":"deepseek-ai/DeepSeek-V3.2","publisher":"Hugging Face","type":"docs","retrieved":"2026-08-20"}]}
{"id":"2025-12-01-runway-gen-4-5","date":"2025-12-01","precision":"day","title":"Runway ships Gen-4.5, its sharpest video model yet","summary":"A new flagship video generation model with tighter prompt adherence and better physics: human movement, camera dynamics, cause and effect. It topped the Artificial Analysis text-to-video leaderboard on release.","category":"model","orgs":["Runway"],"tier":3,"source":{"url":"https://runway.com/research/introducing-runway-gen-4.5","title":"Runway Gen-4.5: State-of-the-Art AI Video Generation","publisher":"Runway","type":"announcement","date":"2025-12-01"},"references":[{"url":"https://mlq.ai/news/runway-launches-gen-45-ai-video-model-offering-hd-text-to-video-generation-with-advanced-physics/","title":"Runway Launches Gen 4.5 AI Video Model Offering HD Text-to-Video Generation with Advanced Physics","publisher":"MLQ.ai","type":"news","date":"2025-12-01"}]}
{"id":"2025-12-02-amazon-nova-2","date":"2025-12-02","precision":"day","title":"Amazon launches its Nova 2 model family at re:Invent","summary":"Nova 2 Lite ships generally available with adjustable reasoning depth and a million token context window, while Nova 2 Pro enters preview. Amazon paired the launch with Trainium3 UltraServers and on-site AI Factories for its cloud customers.","category":"model","orgs":["Amazon"],"tier":2,"source":{"url":"https://aws.amazon.com/blogs/aws/introducing-amazon-nova-2-lite-a-fast-cost-effective-reasoning-model/","title":"Introducing Amazon Nova 2 Lite, a fast, cost-effective reasoning model","publisher":"AWS","type":"announcement","date":"2025-12-02"},"references":[{"url":"https://www.aboutamazon.com/news/aws/aws-re-invent-2025-ai-news-updates","title":"AWS re:Invent 2025: Amazon announces Nova 2, Trainium3, frontier agents","publisher":"About Amazon","type":"news","date":"2025-12-02"}]}
{"id":"2025-12-02-anthropic-acquires-bun","date":"2025-12-02","precision":"day","title":"Anthropic buys Bun","summary":"The JavaScript runtime that Claude Code already runs on went to the company that ships Claude Code, with Bun staying open source and MIT licensed and the same team on it. Anthropic said it is betting on Bun as the infrastructure under its coding products.","category":"acquisition","orgs":["Anthropic","Bun"],"tier":2,"source":{"url":"https://bun.com/blog/bun-joins-anthropic","title":"Bun is joining Anthropic","publisher":"Bun","type":"announcement","date":"2025-12-02"}}
{"id":"2025-12-02-mistral-3","date":"2025-12-02","precision":"day","title":"Mistral ships its third family under Apache 2.0","summary":"Mistral Large 3, a 675 billion parameter mixture of experts with 41 billion active, alongside Ministral models at 14B, 8B and 3B in base, instruct and reasoning variants. Mistral says Large 3 entered LMArena at number two among open non-reasoning models.","category":"model","orgs":["Mistral AI"],"tier":2,"source":{"url":"https://mistral.ai/news/mistral-3","title":"Introducing Mistral 3","publisher":"Mistral AI","type":"announcement","date":"2025-12-02"}}
{"id":"2025-12-02-openai-code-red","date":"2025-12-02","precision":"day","title":"OpenAI calls a code red over Gemini","summary":"Two weeks after Gemini 3 shipped, Altman told staff ChatGPT was at a critical time and pushed the advertising plans back to make room for fixing it. The memo warned of temporary economic headwinds and rough vibes.","category":"company","orgs":["OpenAI","Google"],"tier":2,"source":{"url":"https://fortune.com/2025/12/02/sam-altman-declares-code-red-google-gemini-ceo-sundar-pichai","title":"Sam Altman declares 'Code Red' as Google's Gemini surges—three years after ChatGPT caused Google CEO Sundar Pichai to do the same","publisher":"Fortune","author":"Sharon Goldman","type":"news","date":"2025-12-02"}}
{"id":"2025-12-02-zig-leaves-github","date":"2025-12-02","precision":"day","title":"Zig leaves GitHub over what AI did to it","summary":"The Zig Software Foundation moved to Codeberg, with its president saying GitHub Actions had started picking jobs to run at random after GitHub's chief executive told staff to embrace AI or get out. The Dillo browser project announced the same move.","category":"culture","orgs":["Zig Software Foundation","GitHub"],"tier":3,"source":{"url":"https://www.theregister.com/2025/12/02/zig_quits_github_microsoft_ai_obsession/","title":"Zig quits GitHub, says Microsoft's AI obsession has ruined the service","publisher":"The Register","author":"Thomas Claburn","type":"news","date":"2025-12-02"}}
{"id":"2025-12-04-anthropic-interviewer","date":"2025-12-04","precision":"day","title":"Anthropic builds an AI that interviews workers about AI","summary":"A tool that drafts questions, conducts interviews and analyzes responses at scale, piloted on 1,250 professionals across the general workforce, creative fields and science. Anthropic used the results to publish its first read on how work is actually changing.","category":"research","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/research/anthropic-interviewer","title":"Introducing Anthropic Interviewer: What 1,250 professionals told us about working with AI","publisher":"Anthropic","type":"announcement","date":"2025-12-04"},"references":[{"url":"https://dataconomy.com/2025/12/05/anthropic-launches-interviewer-bot-to-quiz-users-on-ai-usage/","title":"Anthropic Launches Interviewer Bot To Quiz Users On AI Usage","publisher":"Dataconomy","type":"news","date":"2025-12-05"}]}
{"id":"2025-12-04-titans-miras","date":"2025-12-04","precision":"day","title":"Google gives a model a memory that grows","summary":"Titans replaces the fixed-size vector an RNN carries with a deep network that learns what to keep, and MIRAS is the framework describing the design space around it. The target is context lengths a transformer cannot afford to attend over.","category":"research","orgs":["Google"],"tier":3,"source":{"url":"https://research.google/blog/titans-miras-helping-ai-have-long-term-memory/","title":"Titans + MIRAS: Helping AI have long-term memory","publisher":"Google","author":"Ali Behrouz, Meisam Razaviyayn and Vahab Mirrokni","type":"announcement","date":"2025-12-04"}}
{"id":"2025-12-08-ibm-acquires-confluent","date":"2025-12-08","precision":"day","title":"IBM agrees to buy Confluent for $11 billion","summary":"IBM will pay $31 a share for the company that commercialized Apache Kafka, betting that enterprise AI agents need a real-time data layer as much as they need models. The deal is expected to close by mid-2026.","category":"acquisition","orgs":["IBM","Confluent"],"tier":2,"source":{"url":"https://newsroom.ibm.com/2025-12-08-ibm-to-acquire-confluent-to-create-smart-data-platform-for-enterprise-generative-ai","title":"IBM to Acquire Confluent to Create Smart Data Platform for Enterprise Generative AI","publisher":"IBM","type":"press-release","date":"2025-12-08"},"references":[{"url":"https://www.confluent.io/blog/ibm-to-acquire-confluent/","title":"IBM to Acquire Confluent","publisher":"Confluent","type":"announcement","date":"2025-12-08"}]}
{"id":"2025-12-08-nvidia-h200-china-exports","date":"2025-12-08","precision":"day","title":"Trump clears Nvidia to sell H200 chips into China","summary":"A reversal of the export limits the Biden administration imposed, letting approved Chinese customers buy Nvidia's H200 in exchange for 25 percent of the sales going to the US government. Nvidia's newer Blackwell chips stay restricted.","category":"policy","orgs":["NVIDIA","White House"],"tier":2,"source":{"url":"https://www.aljazeera.com/economy/2025/12/9/trump-clears-way-for-sale-of-powerful-nvidia-h200-chips-to-china","title":"Trump clears way for sale of powerful Nvidia H200 chips to China","publisher":"Al Jazeera","type":"news","date":"2025-12-09"},"references":[{"url":"https://www.cfr.org/expert-brief/consequences-exporting-nvidias-h200-chips-china","title":"The Consequences of Exporting Nvidia's H200 Chips to China","publisher":"Council on Foreign Relations","type":"analysis"}]}
{"id":"2025-12-09-agentic-ai-foundation","date":"2025-12-09","precision":"day","title":"OpenAI, Anthropic and Block put their agent standards under one roof","summary":"The Linux Foundation launched the Agentic AI Foundation, taking in Anthropic's Model Context Protocol, Block's goose framework and OpenAI's AGENTS.md as its first projects. Amazon, Google, Microsoft, Bloomberg and Cloudflare joined as platinum members.","category":"open-source","orgs":["Linux Foundation","Anthropic","OpenAI","Block"],"tier":2,"source":{"url":"https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation","title":"Linux Foundation Announces the Formation of the Agentic AI Foundation (AAIF), Anchored by New Project Contributions Including Model Context Protocol (MCP), goose and AGENTS.md","publisher":"Linux Foundation","type":"press-release","date":"2025-12-09"},"references":[{"url":"https://techcrunch.com/2025/12/09/openai-anthropic-and-block-join-new-linux-foundation-effort-to-standardize-the-ai-agent-era/","title":"OpenAI, Anthropic, and Block join new Linux Foundation effort to standardize the AI agent era","publisher":"TechCrunch","type":"news","date":"2025-12-09"},{"url":"https://siliconangle.com/2025/12/09/linux-foundation-announces-agentic-ai-foundation-joined-anthropic-openai-block/","title":"Linux Foundation announces Agentic AI Foundation joined by Anthropic, OpenAI, Block","publisher":"SiliconANGLE","type":"news","date":"2025-12-09"}]}
{"id":"2025-12-09-mistral-devstral-2","date":"2025-12-09","precision":"day","title":"Mistral ships Devstral 2 and a local-first coding CLI","summary":"Devstral 2 is a 123 billion parameter coding model under a modified MIT license, paired with Mistral Vibe, a command line tool that can run entirely on a developer's own machine. Mistral says the code never has to leave it.","category":"model","orgs":["Mistral AI"],"tier":3,"source":{"url":"https://mistral.ai/news/devstral-2-vibe-cli/","title":"Introducing: Devstral 2 and Mistral Vibe CLI","publisher":"Mistral AI","type":"announcement","date":"2025-12-09"},"references":[{"url":"https://mlq.ai/news/mistral-ai-launches-devstral-2-coding-model-and-mistral-vibe-cli-for-vibe-coding-workflows/","title":"Mistral AI launches Devstral 2 coding model and Mistral Vibe CLI for 'vibe coding' workflows","publisher":"MLQ.ai","type":"news","date":"2025-12-09"}]}
{"id":"2025-12-11-ai-preemption-executive-order","date":"2025-12-11","precision":"day","title":"An executive order takes aim at state AI laws","summary":"The order directed federal agencies to challenge state AI regulation and to condition some federal funding on it, stopping short of express preemption. Every state legislature had spent the year writing rules the administration wanted replaced by one.","category":"policy","orgs":["White House"],"tier":2,"source":{"url":"https://www.whitehouse.gov/presidential-actions/2025/12/eliminating-state-law-obstruction-of-national-artificial-intelligence-policy/","title":"Ensuring a National Policy Framework for Artificial Intelligence","publisher":"The White House","type":"press-release","date":"2025-12-11"},"references":[{"url":"https://www.gibsondunn.com/president-trump-latest-executive-order-on-ai-seeks-to-preempt-state-laws/","title":"President Trump's Latest Executive Order on AI Seeks to Preempt State Laws","publisher":"Gibson Dunn","type":"analysis","date":"2025-12-15"}]}
{"id":"2025-12-11-disney-openai-sora-deal","date":"2025-12-11","precision":"day","title":"Disney licenses its characters into Sora","summary":"A three-year agreement putting more than 200 characters from Marvel, Pixar, and Star Wars into Sora and ChatGPT Images, alongside a $1 billion investment in OpenAI. Three months later OpenAI shut Sora down and the deal went with it.","category":"culture","orgs":["The Walt Disney Company","OpenAI"],"tier":2,"source":{"url":"https://thewaltdisneycompany.com/news/disney-openai-sora-agreement/","title":"The Walt Disney Company and OpenAI Reach Agreement to Bring Disney Characters to Sora","publisher":"The Walt Disney Company","type":"press-release","date":"2025-12-11"}}
{"id":"2025-12-11-gpt-5-2","date":"2025-12-11","precision":"day","title":"GPT-5.2 answers Gemini 3 in three weeks","summary":"OpenAI's reply to Google's release, four months after GPT-5 itself. Three more point releases followed within seven months, at which point the version number stopped carrying much information.","category":"model","orgs":["OpenAI"],"tier":3,"source":{"url":"https://openai.com/index/introducing-gpt-5-2/","title":"Introducing GPT-5.2","publisher":"OpenAI","type":"announcement","date":"2025-12-11"}}
{"id":"2025-12-11-time-person-of-the-year","date":"2025-12-11","precision":"day","title":"Time names the architects of AI","summary":"Person of the Year went to a group rather than a person: the people who imagined, built, and financed the buildout. The magazine had done this for the personal computer in 1982 and for \"You\" in 2006.","category":"culture","orgs":["TIME"],"tier":3,"source":{"url":"https://time.com/7339621/person-of-the-year-2025-ai-architects-choice/","title":"Why the Architects of AI Are TIME's 2025 Person of the Year","publisher":"TIME","author":"Sam Jacobs","type":"announcement","date":"2025-12-11"}}
{"id":"2025-12-15-us-tech-force","date":"2025-12-15","precision":"day","title":"The federal government stands up a Tech Force to hire AI talent","summary":"OPM launched a two-year fellowship recruiting about 1,000 early-career engineers and data scientists into federal agencies to modernize systems and build AI capability, with alumni promised hiring consideration at partner companies afterward.","category":"policy","orgs":["White House"],"tier":3,"source":{"url":"https://www.opm.gov/news/news-releases/opm-launches-us-tech-force-to-implement-president-trumps-vision-for-technology-leadership/","title":"OPM Launches US Tech Force to Implement President Trump's Vision for Technology Leadership","publisher":"U.S. Office of Personnel Management","type":"press-release","date":"2025-12-15"},"references":[{"url":"https://fedscoop.com/opm-launches-tech-force-recruit-technologists-government/","title":"OPM launches Tech Force to recruit technologists to government","publisher":"FedScoop","type":"news","date":"2025-12-15"}]}
{"id":"2025-12-15-vpn-extensions-sell-ai-chats","date":"2025-12-15","precision":"day","title":"Privacy extensions were selling their users' AI conversations","summary":"Eight Chrome and Edge extensions sold as privacy tools, Urban VPN among them, had been capturing prompts and responses from ChatGPT, Claude, Gemini, Copilot and five others since July and passing them to an affiliated data broker. More than 8 million installs between them.","category":"research","orgs":["Koi Security"],"tier":3,"source":{"url":"https://www.koi.ai/blog/urban-vpn-browser-extension-ai-conversations-data-collection","title":"8 Million Users' AI Conversations Sold for Profit by 'Privacy' Extensions","publisher":"Koi Security","author":"Idan Dardikman","type":"report","date":"2025-12-15"}}
{"id":"2025-12-16-gpt-image-1-5","date":"2025-12-16","precision":"day","title":"OpenAI ships GPT Image 1.5, four times faster than its predecessor","summary":"Built directly into the GPT-5 architecture rather than a separate diffusion system, the new image model edits faster, follows instructions more precisely, and costs 20 percent less per image token. It shipped a week after Google's Nano Banana Pro forced the comparison.","category":"model","orgs":["OpenAI"],"tier":3,"source":{"url":"https://techcrunch.com/2025/12/16/openai-continues-on-its-code-red-warpath-with-new-image-generation-model/","title":"OpenAI continues on its 'code red' warpath with new image generation model","publisher":"TechCrunch","type":"news","date":"2025-12-16"},"references":[{"url":"https://openai.com/index/new-chatgpt-images-is-here/","title":"The new ChatGPT Images is here","publisher":"OpenAI","type":"announcement"},{"url":"https://www.pcworld.com/article/3014153/openai-launches-gpt-image-1-5-with-faster-generation-and-smarter-edits.html","title":"OpenAI launches GPT Image 1.5 with faster generation and smarter edits","publisher":"PCWorld","type":"news","date":"2025-12-17"}]}
{"id":"2025-12-17-gemini-3-flash","date":"2025-12-17","precision":"day","title":"Gemini 3 Flash pushes the frontier down the price curve","summary":"The cheap tier of Gemini 3, a month after the flagship, aimed at work nobody would have sent to a frontier model at frontier prices. The interesting number in a model release had become dollars per million tokens.","category":"model","orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://blog.google/products/gemini/gemini-3-flash/","title":"Gemini 3 Flash: Frontier intelligence built for speed","publisher":"Google","type":"announcement","date":"2025-12-17"}}
{"id":"2025-12-18-chatgpt-app-directory","date":"2025-12-18","precision":"day","title":"ChatGPT opens an app store to outside developers","summary":"The App Directory went live for every Plus, Team and Enterprise user, with Spotify, DoorDash, Zillow and Khan Academy already listed and public submissions now open. ChatGPT becomes a platform other companies build inside rather than just a product people visit.","category":"product","orgs":["OpenAI"],"tier":2,"source":{"url":"https://www.techcrunch.com/2025/12/18/chatgpt-launches-an-app-store-lets-developers-know-its-open-for-business","title":"ChatGPT launches an app store, lets developers know it's open for business","publisher":"TechCrunch","type":"news","date":"2025-12-18"},"references":[{"url":"https://venturebeat.com/technology/openai-now-accepting-chatgpt-app-submissions-from-third-party-devs-launches","title":"OpenAI now accepting ChatGPT app submissions from third-party devs, launches App Directory","publisher":"VentureBeat","type":"news"}]}
{"id":"2025-12-22-flock-cameras-exposed","date":"2025-12-22","precision":"day","title":"Flock's AI surveillance cameras were livestreaming to the open internet","summary":"At least 60 of Flock's Condor cameras had no password on their admin panels, giving anyone 30 days of archived footage, live feeds of parks and streets, and the ability to change settings. Flock called it a configuration issue affecting a small number of devices.","category":"research","orgs":["Flock Safety"],"tier":3,"source":{"url":"https://www.404media.co/flock-exposed-its-ai-powered-cameras-to-the-internet-we-tracked-ourselves/","title":"Flock Exposed Its AI-Powered Cameras to the Internet. We Tracked Ourselves","publisher":"404 Media","type":"news","date":"2025-12-22"},"references":[{"url":"https://petapixel.com/2025/12/29/big-brother-left-the-door-open-flocks-ai-surveillance-cameras-exposed-to-the-internet/","title":"Big Brother Left the Door Open: Flock's AI Surveillance Cameras Exposed to the Internet","publisher":"PetaPixel","type":"news","date":"2025-12-29"}]}
{"id":"2025-12-24-nvidia-licenses-groq","date":"2025-12-24","precision":"day","title":"Nvidia buys everything about Groq except Groq","summary":"A non-exclusive licence to Groq's inference technology, plus founder Jonathan Ross, president Sunny Madra and much of the team moving to Nvidia, while Groq stayed a company with a new chief executive. Reported at $20 billion, which neither side confirmed.","category":"deal","orgs":["NVIDIA","Groq"],"tier":2,"source":{"url":"https://groq.com/newsroom/groq-and-nvidia-enter-non-exclusive-inference-technology-licensing-agreement-to-accelerate-ai-inference-at-global-scale","title":"Groq and Nvidia Enter Non-Exclusive Inference Technology Licensing Agreement to Accelerate AI Inference at Global Scale","publisher":"Groq","type":"press-release","date":"2025-12-24"},"references":[{"url":"https://siliconangle.com/2025/12/24/nvidia-license-technology-inference-chip-startup-groq-reported-20b-deal/","title":"Nvidia to license technology from inference chip startup Groq in reported $20B deal","publisher":"SiliconANGLE","type":"news","date":"2025-12-24"}]}
{"id":"2025-12-30-meta-acquires-manus","date":"2025-12-30","precision":"day","title":"Meta buys Manus for more than $2 billion","summary":"The Singapore agent company had gone from launch to more than $100 million of annual recurring revenue in eight months. Meta's condition was no continuing Chinese ownership and a wind-down of Manus operations in China, which is the thread Beijing pulled on four months later.","category":"acquisition","orgs":["Meta","Manus"],"tier":2,"source":{"url":"https://www.euronews.com/next/2025/12/31/meta-to-acquire-ai-startup-manus-in-deal-valued-at-over-2-billion","title":"Meta to acquire AI startup Manus in deal valued at over $2 billion","publisher":"Euronews","author":"Theo Farrant","type":"news","date":"2025-12-31"}}
{"id":"2026-01-05-aws-raises-gpu-prices","date":"2026-01-05","precision":"day","title":"AWS raises GPU rental prices by 15 percent","summary":"EC2 Capacity Blocks for ML went up about 15 percent over a weekend, taking the p5e.48xlarge from $34.61 to $39.80 an hour in most regions. AWS put it down to the supply and demand patterns it expected that quarter.","category":"product","orgs":["Amazon"],"tier":3,"source":{"url":"https://www.theregister.com/2026/01/05/aws_price_increase/","title":"AWS raises GPU prices 15% on a Saturday, hopes you weren't paying attention","publisher":"The Register","author":"Corey Quinn","type":"news","date":"2026-01-05"}}
{"id":"2026-01-05-nvidia-rubin-full-production","date":"2026-01-05","precision":"day","title":"NVIDIA puts the six-chip Rubin platform into full production","summary":"Jensen Huang said Rubin, the successor to Blackwell, is shipping in volume, with Vera Rubin racks cutting the cost per token to roughly a tenth of the prior platform. Cloud partners including AWS, Google Cloud, Microsoft and Oracle are due to deploy it later in 2026.","category":"hardware","orgs":["NVIDIA"],"tier":2,"source":{"url":"https://blogs.nvidia.com/blog/2026-ces-special-presentation/","title":"NVIDIA Rubin Platform, Open Models, Autonomous Driving: NVIDIA Presents Blueprint for the Future at CES","publisher":"NVIDIA","type":"announcement","date":"2026-01-05"},"references":[{"url":"https://www.servethehome.com/nvidia-launches-next-generation-rubin-ai-compute-platform-at-ces-2026/","title":"NVIDIA Launches Next-Generation Rubin AI Compute Platform at CES 2026","publisher":"ServeTheHome","type":"news"},{"url":"https://videocardz.com/newz/nvidia-vera-rubin-nvl72-detailed-72-gpus-36-cpus-260-tb-s-scale-up-bandwidth","title":"NVIDIA Vera Rubin NVL72 Detailed: 72 GPUs, 36 CPUs, 260 TB/s Scale-Up Bandwidth","publisher":"VideoCardz","type":"news"},{"url":"https://www.datacenterdynamics.com/en/news/nvidia-ceo-announces-vera-rubin-chips-are-in-full-production-during-ces-keynote/","title":"Nvidia CEO announces Vera Rubin chips are in full production during CES keynote","publisher":"Data Center Dynamics","type":"news"}]}
{"id":"2026-01-07-chatgpt-health","date":"2026-01-07","precision":"day","title":"OpenAI opens a health wing inside ChatGPT","summary":"A separate space for health questions, wired to Apple Health, Function and MyFitnessPal so answers sit on the user's own data. OpenAI said 230 million people already ask it health questions every week and that these conversations would not be used to train models.","category":"product","orgs":["OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2026/01/07/openai-unveils-chatgpt-health-says-230-million-users-ask-about-health-each-week/","title":"OpenAI unveils ChatGPT Health, says 230 million users ask about health each week","publisher":"TechCrunch","author":"Amanda Silberling","type":"news","date":"2026-01-07"}}
{"id":"2026-01-08-zhipu-ai-hong-kong-listing","date":"2026-01-08","precision":"day","title":"Zhipu AI lists on the Hong Kong exchange","summary":"Zhipu AI, now branded internationally as Z.ai, began trading on HKEX's Main Board as 02513. The company raised HK$4.3 billion in the offering, giving public-market investors a direct stake in a Chinese foundation-model developer.","category":"company","orgs":["Zhipu AI"],"tier":2,"source":{"url":"https://www.cyberport.hk/file/Press_release/Jan%202026/%5BPress%20Release%5D%20Cyberport%20Enterprise%20Z.ai%20Debuts%20on%20HKEX%20as%20the%20%E2%80%9CWorld%E2%80%99s%20First%20Large%20Model%20Stock%E2%80%9D%20_final.pdf","title":"Cyberport Enterprise Z.ai Debuts on HKEX","publisher":"Cyberport","type":"press-release","date":"2026-01-08"},"references":[{"url":"https://www.hkexgroup.com/Media-Centre/Insight/Insight/2026/Johnson-Chui/AI-IPOs-Drive-a-Strong-Start-to-2026?sc_lang=en","title":"AI IPOs Drive a Strong Start to 2026","publisher":"HKEX Group","type":"analysis","date":"2026-01-16"}]}
{"id":"2026-01-12-apple-licenses-gemini","date":"2026-01-12","precision":"day","title":"Apple bases its next foundation models on Gemini","summary":"Apple and Google entered a multi-year collaboration under which the next Apple Foundation Models would be based on Gemini models and Google cloud technology. Apple said its intelligence features would still run on its devices and Private Cloud Compute.","category":"company","orgs":["Apple","Google"],"tier":1,"source":{"url":"https://blog.google/company-news/inside-google/company-announcements/joint-statement-google-apple/","title":"Joint statement from Google and Apple","publisher":"Google","type":"announcement","date":"2026-01-12"},"references":[{"url":"https://www.cnbc.com/2026/01/12/apple-google-ai-siri-gemini.html","title":"Apple picks Google's Gemini to run AI-powered Siri coming this year","publisher":"CNBC","type":"news","date":"2026-01-12"},{"url":"https://www.theantitrustattorney.com/apples-gemini-siri-deal-is-the-next-microsoft-antitrust-case-not-the-next-app-store-fight/","title":"Apple's Gemini-Siri Deal Is the Next Microsoft Antitrust Case, Not the Next App Store Fight","publisher":"The Antitrust Attorney","author":"Luis Blanquez","type":"analysis","date":"2026-05-14"}]}
{"id":"2026-01-12-claude-cowork","date":"2026-01-12","precision":"day","title":"Claude Cowork takes the agent out of the terminal","summary":"Anthropic released Cowork as a Mac research preview for delegating multi-step work over selected folders without opening a terminal. Two days later, a security vendor demonstrated how a malicious document and an allowlisted network route could expose those files.","category":"product","orgs":["Anthropic"],"tier":2,"source":{"url":"https://claude.com/blog/cowork-research-preview","title":"Cowork: Claude Code for the rest of your work","publisher":"Anthropic","type":"announcement","date":"2026-01-12","archiveUrl":"https://web.archive.org/web/20260112194131/https://claude.com/blog/cowork-research-preview"},"references":[{"url":"https://simonwillison.net/2026/Jan/12/claude-cowork/","title":"First impressions of Claude Cowork, Anthropic's general agent","publisher":"Simon Willison","author":"Simon Willison","type":"analysis","date":"2026-01-12"}]}
{"id":"2026-01-12-erdos-problem-728","date":"2026-01-12","precision":"day","title":"An Erdős problem falls to a machine","summary":"GPT-5.2 Pro produced the informal argument, Harmonic's Aristotle turned it into a Lean proof, and Kevin Barreto drove the pair. The writeup calls it the first Erdős problem regarded as fully resolved autonomously by an AI system.","category":"research","orgs":["OpenAI","Harmonic"],"tier":2,"source":{"url":"https://arxiv.org/abs/2601.07421","title":"Resolution of Erdős Problem #728: a writeup of Aristotle's Lean proof","publisher":"arXiv","author":"Nat Sothanaphan","type":"paper","date":"2026-01-12"}}
{"id":"2026-01-13-anthropic-python-software-foundation","date":"2026-01-13","precision":"day","title":"Anthropic puts $1.5 million into the Python Software Foundation","summary":"Two years of funding aimed at the security work that keeps supply chain attacks off PyPI, along with the Developer in Residence program, community grants and core infrastructure.","category":"company","orgs":["Anthropic","Python Software Foundation"],"tier":3,"source":{"url":"https://discuss.python.org/t/anthropic-has-made-a-large-contribution-to-the-python-software-foundation-and-open-source-security/105694","title":"Anthropic has made a large contribution to the Python Software Foundation and open source security","publisher":"Python Software Foundation","type":"announcement","date":"2026-01-13"}}
{"id":"2026-01-13-h200-china-export-policy","date":"2026-01-13","precision":"day","title":"Commerce Department opens a licensing path for H200 sales to China","summary":"BIS moved Nvidia's H200 and AMD's MI325X to case by case export review for China, requiring buyer compliance screening and independent security testing rather than a blanket ban, following Trump's December pledge to allow the sales for a cut of the revenue.","category":"policy","orgs":["NVIDIA","AMD","United States Department of Commerce"],"tier":2,"source":{"url":"https://www.bis.gov/press-release/department-commerce-revises-license-review-policy-semiconductors-exported-china","title":"Department of Commerce Revises License Review Policy for Semiconductors Exported to China","publisher":"US Department of Commerce, Bureau of Industry and Security","type":"press-release","date":"2026-01-13"},"references":[{"url":"https://malaymail.com/news/tech-gadgets/2026/01/14/us-allows-nvidia-to-sell-h200-ai-chips-to-china-but-top-models-remain-restricted/205446","title":"US allows Nvidia to sell H200 AI chips to China but top models remain restricted","publisher":"Malay Mail","type":"news","date":"2026-01-14"},{"url":"https://www.nextgov.com/emerging-tech/2026/01/lawmakers-worry-over-new-rule-will-allow-sales-nvidias-h200-chips-china/410696/","title":"Lawmakers worry over new rule that will allow sales of Nvidia's H200 chips to China","publisher":"Nextgov/FCW","type":"news"}]}
{"id":"2026-01-14-cowork-file-exfiltration","date":"2026-01-14","precision":"day","title":"Claude Cowork is made to hand over your files","summary":"Cowork only allows outbound traffic to an allowlist, and Anthropic's own API is on it. PromptArmor hid instructions in a document that had the agent upload every file it could see to an attacker's Anthropic account, with no approval step anywhere, two days after Cowork shipped.","category":"research","orgs":["Anthropic","PromptArmor"],"tier":2,"source":{"url":"https://simonwillison.net/2026/Jan/14/claude-cowork-exfiltrates-files/","title":"Claude Cowork Exfiltrates Files","publisher":"Simon Willison","author":"Simon Willison","type":"analysis","date":"2026-01-14"},"references":[{"url":"https://www.promptarmor.com/resources/claude-cowork-exfiltrates-files","title":"Claude Cowork Exfiltrates Files","publisher":"PromptArmor","type":"report","retrieved":"2026-08-20"}]}
{"id":"2026-01-14-openai-cerebras-inference-deal","date":"2026-01-14","precision":"day","title":"OpenAI signs a multi-year inference deal with Cerebras","summary":"OpenAI committed to 750 megawatts of Cerebras wafer-scale chips for low-latency inference, in a deal Cerebras says is worth more than $10 billion, its largest customer contract yet and a hedge against reliance on the United Arab Emirates' G42.","category":"deal","orgs":["OpenAI","Cerebras"],"tier":2,"source":{"url":"https://www.cerebras.ai/blog/openai-partners-with-cerebras-to-bring-high-speed-inference-to-the-mainstream","title":"OpenAI Partners with Cerebras to Bring High-Speed Inference to the Mainstream","publisher":"Cerebras","type":"announcement","date":"2026-01-14"},"references":[{"url":"https://techcrunch.com/2026/01/14/openai-signs-deal-reportedly-worth-10-billion-for-compute-from-cerebras/","title":"OpenAI signs deal reportedly worth $10 billion for compute from Cerebras","publisher":"TechCrunch","type":"news","date":"2026-01-14"},{"url":"https://www.theregister.com/2026/01/15/openai_cerebras_ai/","title":"OpenAI to serve up ChatGPT on Cerebras' AI dinner plates in $10B+ deal","publisher":"The Register","type":"news","date":"2026-01-15"}]}
{"id":"2026-01-14-thinking-machines-lab-loses-cofounders","date":"2026-01-14","precision":"day","title":"Thinking Machines Lab loses two co-founders back to OpenAI","summary":"Mira Murati said Thinking Machines Lab had parted ways with CTO Barret Zoph. OpenAI announced within the hour that Zoph, fellow co-founder Luke Metz and researcher Sam Schoenholz were returning, less than a year after the lab's record seed round.","category":"company","orgs":["Thinking Machines Lab","OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2026/01/14/mira-muratis-startup-thinking-machines-lab-is-losing-two-of-its-co-founders-to-openai","title":"Mira Murati's startup, Thinking Machines Lab, is losing two of its co-founders to OpenAI","publisher":"TechCrunch","author":"Lucas Ropek","type":"news","date":"2026-01-14"},"references":[{"url":"https://brendonbeebe.substack.com/p/thinking-machines-lab-timeline-of","title":"Thinking Machines Lab: Timeline of the $2B Seed Round, Tinker, and the January 2026 OpenAI Exodus","publisher":"Brendon Beebe","type":"retrospective","date":"2026-01-18"}]}
{"id":"2026-01-16-clickhouse-acquires-langfuse","date":"2026-01-16","precision":"day","title":"ClickHouse buys Langfuse","summary":"The open source platform teams point their LLM applications at for tracing and evaluation went to the database company whose engine it already ran on, with the product staying open source and self-hostable. Langfuse's founders said they had term sheets for a Series A and took this instead.","category":"acquisition","orgs":["ClickHouse","Langfuse"],"tier":3,"source":{"url":"https://langfuse.com/blog/joining-clickhouse","title":"Langfuse joins ClickHouse","publisher":"Langfuse","type":"announcement","date":"2026-01-16"}}
{"id":"2026-01-18-xai-colossus-2-gigawatt-claim","date":"2026-01-18","precision":"day","title":"xAI says Colossus 2 becomes the first gigawatt AI training cluster","summary":"Elon Musk said the supercomputer built for Grok training was running at gigawatt scale, with an upgrade to 1.5 gigawatts due in April. Independent satellite imagery analysis put the site's cooling capacity at roughly a third of that figure.","category":"hardware","orgs":["xAI"],"tier":3,"source":{"url":"https://www.teslarati.com/elon-musk-xai-brings-1gw-colossus-2-ai-training-cluster-online/","title":"Elon Musk's xAI brings 1GW Colossus 2 AI training cluster online","publisher":"Teslarati","author":"Simon Alvarez","type":"news","date":"2026-01-18"},"references":[{"url":"https://www.tomshardware.com/tech-industry/artificial-intelligence/elon-musks-xai-colossus-2-is-nowhere-near-1-gigawatt-capacity-satellite-imagery-suggests-despite-claims-site-only-has-350-megawatts-of-cooling-capacity","title":"Elon Musk's xAI Colossus 2 is nowhere near 1 gigawatt capacity, satellite imagery suggests — despite claims, site only has 350 megawatts of cooling capacity","publisher":"Tom's Hardware","type":"news"}]}
{"id":"2026-01-21-gptzero-neurips-hallucinations","date":"2026-01-21","precision":"day","title":"Fabricated citations turn up in NeurIPS papers","summary":"GPTZero went through all 4,841 papers NeurIPS accepted in 2025 and found 100 confirmed hallucinations spread across 53 of them, mostly citations to work that does not exist. All of it had cleared peer review.","category":"research","orgs":["GPTZero"],"tier":3,"source":{"url":"https://gptzero.me/news/neurips","title":"GPTZero finds 100 new hallucinations in NeurIPS 2025 accepted papers","publisher":"GPTZero","type":"report","date":"2026-01-21"}}
{"id":"2026-01-22-claude-constitution","date":"2026-01-22","precision":"day","title":"Anthropic publishes Claude's constitution","summary":"A document describing the values and judgment Anthropic wants Claude to have, written to be read by Claude rather than by users, and used to shape training. It went out under CC0, so anyone can take it and use it.","category":"company","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/news/claude-new-constitution","title":"Claude's new constitution","publisher":"Anthropic","type":"announcement","date":"2026-01-22"}}
{"id":"2026-01-26-microsoft-maia-200","date":"2026-01-26","precision":"day","title":"Microsoft unveils its second-generation Maia AI chip","summary":"Maia 200 packs more than 100 billion transistors and roughly 10 petaflops of 4-bit compute, with Microsoft claiming three times the performance of Amazon's Trainium on some benchmarks. The chip is already handling GPT-5.2 inference and Microsoft 365 Copilot workloads in US data centers.","category":"hardware","orgs":["Microsoft"],"tier":2,"source":{"url":"https://news.microsoft.com/maia-200/","title":"Introducing Microsoft's next-gen AI accelerator: Maia 200","publisher":"Microsoft","type":"announcement","date":"2026-01-26"},"references":[{"url":"https://www.techcrunch.com/2026/01/26/microsoft-announces-powerful-new-chip-for-ai-inference/","title":"Microsoft announces powerful new chip for AI inference","publisher":"TechCrunch","author":"Lucas Ropek","type":"news","date":"2026-01-26"},{"url":"https://www.geekwire.com/2026/microsoft-unveils-maia-200-ai-chip-claiming-performance-edge-over-amazon-and-google/","title":"Microsoft unveils Maia 200 AI chip, claiming performance edge over Amazon and Google","publisher":"GeekWire","type":"news"}]}
{"id":"2026-01-26-nvidia-invests-coreweave","date":"2026-01-26","precision":"day","title":"NVIDIA puts another $2 billion into CoreWeave","summary":"NVIDIA bought CoreWeave stock at $87.20 a share, deepening a circular relationship in which its biggest cloud GPU customer is also one of its own investments. CoreWeave gets early access to new NVIDIA chips and help funding more than 5 gigawatts of AI data centers by 2030.","category":"funding","orgs":["NVIDIA","CoreWeave"],"tier":2,"source":{"url":"https://nvidianews.nvidia.com/news/nvidia-and-coreweave-strengthen-collaboration-to-accelerate-buildout-of-ai-factories","title":"NVIDIA and CoreWeave Strengthen Collaboration to Accelerate Buildout of AI Factories","publisher":"NVIDIA","type":"press-release","date":"2026-01-26"},"references":[{"url":"https://finance.yahoo.com/news/nvidia-nvda-invests-2b-coreweave-070753277.html","title":"NVIDIA (NVDA) Invests $2B in CoreWeave, Expands AI Data Center Partnership","publisher":"Yahoo Finance","type":"news"},{"url":"https://www.coreweave.com/news/nvidia-and-coreweave-strengthen-collaboration-to-accelerate-buildout-of-ai-factories","title":"NVIDIA and CoreWeave Strengthen Collaboration to Accelerate Buildout of AI Factories","publisher":"CoreWeave","type":"press-release","date":"2026-01-26"}]}
{"id":"2026-01-27-clawdbot-becomes-moltbot","date":"2026-01-27","precision":"day","title":"A trademark complaint renames Clawdbot overnight","summary":"Anthropic decided Clawd sat too close to Claude and asked for the name, so Peter Steinberger's personal agent became Moltbot the same morning, mascot included. Squatters took the freed GitHub and X handles within seconds and stood up typosquat domains and a cloned repository.","category":"company","orgs":["Anthropic","OpenClaw"],"tier":2,"source":{"url":"https://laravel-news.com/clawdbot-rebrands-to-moltbot-after-trademark-request-from-anthropic","title":"Clawdbot Rebrands to Moltbot After Trademark Request From Anthropic","publisher":"Laravel News","author":"Eric L. Barnes","type":"news","date":"2026-01-27"},"references":[{"url":"https://www.malwarebytes.com/blog/threat-intel/2026/01/clawdbots-rename-to-moltbot-sparks-impersonation-campaign","title":"Clawdbot's rename to Moltbot sparks impersonation campaign","publisher":"Malwarebytes","author":"Stefan Dasic","type":"report","date":"2026-01-29"}]}
{"id":"2026-01-27-kimi-k2-5","date":"2026-01-27","precision":"day","title":"Moonshot ships Kimi K2.5 with a swarm mode","summary":"A trillion-parameter mixture of experts trained on 15 trillion mixed visual and text tokens, released open weight, with a mode that lets it direct up to a hundred sub-agents at once. Moonshot says it takes the top score on HLE-Full and beats GPT-5.2 on several benchmarks.","category":"model","orgs":["Moonshot AI"],"tier":2,"source":{"url":"https://siliconangle.com/2026/01/27/moonshot-ai-releases-open-source-kimi-k2-5-model-1t-parameters/","title":"Moonshot AI releases open-source Kimi K2.5 model with 1T parameters","publisher":"SiliconANGLE","author":"Maria Deutscher","type":"news","date":"2026-01-27"},"references":[{"url":"https://www.kimi.ai/blog/kimi-k2-5.html","title":"Kimi K2.5: Visual Agentic Intelligence","publisher":"Moonshot AI","type":"announcement","retrieved":"2026-08-20"}]}
{"id":"2026-01-28-moltbook","date":"2026-01-28","precision":"day","title":"A social network opens that humans may read but not post to","summary":"Matt Schlicht launched Moltbook for AI agents only, and the population went from 37,000 to 1.5 million inside a day. The agents, most of them running Steinberger's software, produced religions, a marketplace for prompt injections, and a document they called the Molt Magna Carta.","category":"culture","orgs":["Moltbook","OpenClaw"],"tier":2,"source":{"url":"https://theconversation.com/moltbook-ai-bots-use-social-network-to-create-religions-and-deal-digital-drugs-but-are-some-really-humans-in-disguise-274895","title":"Moltbook: AI bots use social network to create religions and deal digital drugs – but are some really humans in disguise?","publisher":"The Conversation","author":"David Reid","type":"analysis","date":"2026-02-05"}}
{"id":"2026-01-28-tesla-invests-xai","date":"2026-01-28","precision":"day","title":"Tesla commits $2 billion to Elon Musk's xAI","summary":"Tesla's board proceeded with a $2 billion purchase of xAI Series E preferred stock even after shareholders rejected a nonbinding measure to authorize it in November, with abstentions counted as no votes under Tesla's bylaws. The stake buys under 1 percent of a company last valued near $230 billion.","category":"funding","orgs":["Tesla","xAI"],"tier":2,"source":{"url":"https://techcrunch.com/2026/01/28/tesla-invested-2b-in-elon-musks-xai/","title":"Tesla to invest $2B in Elon Musk's xAI","publisher":"TechCrunch","author":"Kirsten Korosec","type":"news","date":"2026-01-28"},"references":[{"url":"https://www.cnbc.com/2026/01/28/tesla-to-invest-2-billion-in-xai-elon-musks-openai-competitor.html","title":"Tesla to invest $2 billion in xAI, Elon Musk's OpenAI competitor","publisher":"CNBC","type":"news","date":"2026-01-28"},{"url":"https://www.teslarati.com/tesla-announces-massive-investment-xai-2-billion/","title":"Tesla announces massive investment into xAI","publisher":"Teslarati","type":"news"}]}
{"id":"2026-01-29-anthropic-coding-skills-study","date":"2026-01-29","precision":"day","title":"Anthropic measures what AI help costs a learner","summary":"In a randomized study of 52 developers learning Trio, the AI group averaged 50 percent on an immediate quiz versus 67 percent without AI, a 17 percentage-point gap. The completion-time difference was not statistically significant; long-term retention was not measured.","category":"research","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/research/AI-assistance-coding-skills","title":"How AI assistance impacts the formation of coding skills","publisher":"Anthropic","author":"Judy Hanwen Shen and Alex Tamkin","type":"paper","date":"2026-01-29"},"references":[{"url":"https://arxiv.org/abs/2601.20245","title":"How AI Impacts Skill Formation","publisher":"arXiv","type":"paper","date":"2026-01-28"}]}
{"id":"2026-01-29-metr-time-horizon-1-1","date":"2026-01-29","precision":"day","title":"METR revises its time-horizon measurements","summary":"METR rebuilt its measure of the task length frontier models can complete with 50 percent reliability, expanding the suite from 170 to 228 tasks and moving its evaluation infrastructure to Inspect. The rerun changed individual estimates and the apparent pace of progress, showing how much the headline depends on the task distribution.","category":"research","orgs":["METR"],"tier":2,"source":{"url":"https://metr.org/blog/2026-1-29-time-horizon-1-1/","title":"Time Horizon 1.1","publisher":"METR","type":"report","date":"2026-01-29"},"references":[{"url":"https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/","title":"Measuring AI Ability to Complete Long Software Tasks","publisher":"METR","type":"report","date":"2025-03-19"}]}
{"id":"2026-02-02-spacex-acquires-xai","date":"2026-02-02","precision":"day","title":"SpaceX acquires xAI","summary":"SpaceX announced that it had acquired xAI, but the announcement disclosed no price or transaction structure. Contemporary reports valued the combined companies at about $1.25 trillion; CNBC described it as the largest merger on record.","category":"acquisition","orgs":["SpaceX","xAI"],"tier":1,"source":{"url":"https://x.ai/news/xai-joins-spacex","title":"xAI joins SpaceX","publisher":"xAI","type":"announcement","date":"2026-02-02"},"references":[{"url":"https://www.nbcnews.com/business/business-news/elon-musks-spacex-acquires-xai-rcna257121","title":"Elon Musk's SpaceX acquires xAI","publisher":"NBC News","type":"news","date":"2026-02-02"},{"url":"https://www.cnbc.com/2026/02/03/musk-xai-spacex-biggest-merger-ever.html","title":"Musk's xAI, SpaceX combo is the biggest merger of all time, valued at $1.25 trillion","publisher":"CNBC","type":"news","date":"2026-02-03"},{"url":"https://www.cnbc.com/2026/02/11/musk-announces-xai-re-org-following-key-departures-spacex-merger.html","title":"Musk announces xAI re-org following co-founder departures, SpaceX merger","publisher":"CNBC","type":"news","date":"2026-02-11"}]}
{"id":"2026-02-05-claude-opus-4-6","date":"2026-02-05","precision":"day","title":"Claude Opus 4.6 adds a million-token beta and adaptive reasoning","summary":"Anthropic released Opus 4.6 across Claude, its API and cloud platforms. The model added a beta one-million-token context window, adjustable reasoning effort and context compaction, with different prices for prompts longer than 200,000 tokens.","category":"model","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/claude-opus-4-6","title":"Introducing Claude Opus 4.6","publisher":"Anthropic","type":"announcement","date":"2026-02-05"}}
{"id":"2026-02-05-gpt-5-3-codex","date":"2026-02-05","precision":"day","title":"GPT-5.3-Codex ships the same day as Claude Opus 4.6","summary":"OpenAI's coding model landed within hours of Anthropic's flagship, which had stopped being a coincidence. A week later a Cerebras partnership produced a version fast enough to change how the tool felt.","category":"model","orgs":["OpenAI"],"tier":3,"source":{"url":"https://openai.com/index/introducing-gpt-5-3-codex/","title":"Introducing GPT-5.3-Codex","publisher":"OpenAI","type":"announcement","date":"2026-02-05"},"references":[{"url":"https://openai.com/index/introducing-gpt-5-3-codex-spark/","title":"Introducing GPT-5.3-Codex-Spark","publisher":"OpenAI","type":"announcement","date":"2026-02-12"}]}
{"id":"2026-02-05-openai-frontier","date":"2026-02-05","precision":"day","title":"OpenAI launches Frontier","summary":"An enterprise platform for building and running agents that OpenAI describes as coworkers, with onboarding, permissions and feedback loops borrowed from how companies manage people. HP, Oracle, State Farm and Uber were on it at launch.","category":"product","orgs":["OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2026/02/05/openai-launches-a-way-for-enterprises-to-build-and-manage-ai-agents/","title":"OpenAI launches a way for enterprises to build and manage AI agents","publisher":"TechCrunch","author":"Rebecca Szkutak","type":"news","date":"2026-02-05"}}
{"id":"2026-02-06-waymo-world-model","date":"2026-02-06","precision":"day","title":"Waymo builds a world model to drive in","summary":"Waymo introduced a Genie 3-based simulator that generates camera and lidar data for its driving system. Engineers can vary routes, traffic and weather to test rare situations without waiting to encounter them on public roads.","category":"research","orgs":["Waymo"],"tier":3,"source":{"url":"https://waymo.com/blog/2026/02/the-waymo-world-model-a-new-frontier-for-autonomous-driving-simulation","title":"The Waymo World Model: A New Frontier For Autonomous Driving Simulation","publisher":"Waymo","type":"announcement","date":"2026-02-06"}}
{"id":"2026-02-09-ads-in-chatgpt","date":"2026-02-09","precision":"day","title":"OpenAI starts testing ads in ChatGPT","summary":"OpenAI began a U.S. test of labeled sponsored placements for logged-in adults on Free and Go. It said ranking could use the conversation topic, past chats and prior ad interactions, while advertisers received aggregate performance data rather than conversation contents.","category":"product","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/testing-ads-in-chatgpt/","title":"Testing ads in ChatGPT","publisher":"OpenAI","type":"announcement","date":"2026-02-09"},"references":[{"url":"https://openai.com/index/our-approach-to-advertising-and-expanding-access/","title":"Our approach to advertising and expanding access to ChatGPT","publisher":"OpenAI","type":"announcement","date":"2026-01-16"}]}
{"id":"2026-02-11-glm-5","date":"2026-02-11","precision":"day","title":"Z.ai releases the open-weight GLM-5","summary":"Z.ai released GLM-5 under MIT terms, a 744-billion-parameter mixture-of-experts model with 40 billion active parameters, alongside API access and local-deployment instructions. Its reported benchmark comparisons are developer measurements, not independently replicated results.","category":"model","secondaryCategories":["open-source"],"orgs":["Zhipu AI"],"tier":2,"source":{"url":"https://huggingface.co/zai-org/GLM-5/blob/83d08ca96c5049150ffb89dd39343746b179e760/README.md","title":"GLM-5","publisher":"Z.ai","type":"docs","date":"2026-02-11"},"references":[{"url":"https://arxiv.org/abs/2602.15763","title":"GLM-5: from Vibe Coding to Agentic Engineering","publisher":"arXiv","type":"paper","date":"2026-02-17"}]}
{"id":"2026-02-12-anthropic-series-g","date":"2026-02-12","precision":"day","title":"Anthropic raises $30 billion at $380 billion","summary":"A Series G that more than doubled the company's valuation in under a year, with Claude Code's weekly active users cited as the headline number. Private markets had stopped pretending these were startups.","category":"funding","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/news/anthropic-raises-30-billion-series-g-funding-380-billion-post-money-valuation","title":"Anthropic raises $30 billion in Series G funding at $380 billion post-money valuation","publisher":"Anthropic","type":"announcement","date":"2026-02-12"},"references":[{"url":"https://www.cnbc.com/2026/02/12/anthropic-closes-30-billion-funding-round-at-380-billion-valuation.html","title":"Anthropic closes $30 billion funding round at $380 billion valuation","publisher":"CNBC","type":"news","date":"2026-02-12"}]}
{"id":"2026-02-12-gemini-3-deep-think","date":"2026-02-12","precision":"day","title":"Google updates Gemini 3 Deep Think for science and engineering","summary":"Google released an updated Deep Think reasoning mode to AI Ultra subscribers in the Gemini app, with selective early API access. It paired improved benchmark results with examples of mathematical review and generating 3D-printable designs.","category":"model","orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-deep-think/","title":"Gemini 3 Deep Think","publisher":"Google","type":"announcement","date":"2026-02-12"}}
{"id":"2026-02-12-gluon-amplitude-result","date":"2026-02-12","precision":"day","title":"GPT-5.2 helps derive a new result in theoretical physics","summary":"Four physicists and a chat model produced a closed-form expression for single-minus gluon amplitudes, which the field had presumed to vanish. One author called it the first real discovery in theoretical physics made by an AI, with the model as a fifth collaborator rather than the researcher.","category":"research","orgs":["OpenAI"],"tier":3,"source":{"url":"https://arxiv.org/abs/2602.12176","title":"Single-minus gluon tree amplitudes are nonzero","publisher":"arXiv","type":"paper","date":"2026-02-12"},"references":[{"url":"https://phys.org/news/2026-02-chatbot-author-ai-stalled-gluon.html","title":"Can a chatbot be a co-author? AI helps crack a long-stalled gluon amplitude proof","publisher":"Phys.org","author":"Eric Moskowitz","type":"news","date":"2026-02-20"}]}
{"id":"2026-02-14-openclaw-creator-joins-openai","date":"2026-02-14","precision":"day","title":"OpenClaw's creator joins OpenAI","summary":"Peter Steinberger took a job building agents for mainstream users, three months after publishing the personal agent that became the most-starred repository on GitHub. OpenClaw went to a foundation, with OpenAI sponsoring the project rather than absorbing it.","category":"company","orgs":["OpenClaw","OpenAI"],"tier":2,"source":{"url":"https://steipete.me/posts/2026/openclaw","title":"I'm joining OpenAI","publisher":"Peter Steinberger","author":"Peter Steinberger","type":"announcement","date":"2026-02-14"}}
{"id":"2026-02-17-claude-sonnet-4-6","date":"2026-02-17","precision":"day","title":"Claude Sonnet 4.6","summary":"Anthropic introduced Claude Sonnet 4.6 with a million-token context window in beta, bringing that capacity to its mid-tier model.","category":"model","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/claude-sonnet-4-6","title":"Introducing Claude Sonnet 4.6","publisher":"Anthropic","type":"announcement","date":"2026-02-17"}}
{"id":"2026-02-19-gemini-3-1-pro","date":"2026-02-19","precision":"day","title":"Google releases Gemini 3.1 Pro","summary":"Google released Gemini 3.1 Pro in preview across its API, enterprise products, the Gemini app and NotebookLM, with consumer access gated to Pro and Ultra plans. Its 77.1 percent ARC-AGI-2 result used the Thinking High setting and measured abstract puzzle solving, not general task success.","category":"model","orgs":["Google"],"tier":2,"source":{"url":"https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/","title":"Gemini 3.1 Pro","publisher":"Google","type":"announcement","date":"2026-02-19"}}
{"id":"2026-02-20-ggml-joins-hugging-face","date":"2026-02-20","precision":"day","title":"ggml.ai joins Hugging Face","summary":"The team behind llama.cpp, which is how most people run a model on hardware they own, moved to Hugging Face with the libraries staying open source and community-run. Local inference acquired an institution.","category":"acquisition","orgs":["Hugging Face","ggml.ai"],"tier":3,"source":{"url":"https://github.com/ggml-org/llama.cpp/discussions/19759","title":"ggml.ai joins Hugging Face to ensure the long-term progress of Local AI","publisher":"ggml.ai","type":"announcement","date":"2026-02-20"}}
{"id":"2026-02-20-mpa-bytedance-cease-and-desist","date":"2026-02-20","precision":"day","title":"Hollywood sends its first cease-and-desist to an AI company","summary":"The Motion Picture Association told ByteDance to stop training on member studios' films and to block users from generating their characters, after clips of an AI Brad Pitt and Tom Cruise spread across social media. The letter called the infringement a feature rather than a bug and asked for an answer by February 27.","category":"policy","orgs":["Motion Picture Association","ByteDance"],"tier":2,"source":{"url":"https://au.variety.com/2026/film/news/motion-picture-association-bytedance-seedance-letter-33338","title":"Motion Picture Association Pushes ByteDance to Curb Seedance 2.0 AI Infringement","publisher":"Variety","author":"Gene Maddaus","type":"news","date":"2026-02-21"}}
{"id":"2026-02-24-anthropic-rsp-rewrite","date":"2026-02-24","precision":"day","title":"Anthropic rewrites its Responsible Scaling Policy","summary":"Version 3 reorganized the policy around company roadmaps, risk reports every three to six months and conditional competitor commitments. The roadmaps were public goals rather than hard commitments; Anthropic said collective-action limits made the old absolute framework impractical.","category":"company","orgs":["Anthropic"],"tier":2,"source":{"url":"https://time.com/7380854/exclusive-anthropic-drops-flagship-safety-pledge/","title":"Exclusive: Anthropic Drops Flagship Safety Pledge","publisher":"TIME","author":"Billy Perrigo","type":"news","date":"2026-02-24"}}
{"id":"2026-02-26-amodei-refuses-department-of-war","date":"2026-02-26","precision":"day","title":"Amodei refuses the Department of War's terms","summary":"The department would contract only with companies that accept any lawful use, and Anthropic would not give up two carve-outs: mass domestic surveillance and fully autonomous weapons. Amodei published the refusal along with the threats attached to it, including invoking the Defense Production Act.","category":"policy","orgs":["Anthropic","United States Department of War"],"tier":2,"source":{"url":"https://www.anthropic.com/news/statement-department-of-war","title":"Statement from Dario Amodei on our discussions with the Department of War","publisher":"Anthropic","author":"Dario Amodei","type":"announcement","date":"2026-02-26"},"references":[{"url":"https://www.anthropic.com/news/statement-comments-secretary-war","title":"Statement on the comments from Secretary of War Pete Hegseth","publisher":"Anthropic","type":"announcement","date":"2026-02-27"}]}
{"id":"2026-02-26-block-cuts-forty-percent","date":"2026-02-26","precision":"day","title":"Block cuts 40 percent of its staff and says AI is the reason","summary":"More than 10,000 people became fewer than 6,000 in a day, with Dorsey saying the business was strong rather than in trouble and that intelligence tools had changed what it takes to run a company. The stock jumped after hours, and he predicted most companies would reach the same conclusion inside a year.","category":"company","orgs":["Block"],"tier":2,"source":{"url":"https://sfstandard.com/2026/02/26/block-lays-off-staff/","title":"AI made him do it: Jack Dorsey lays off 40% of Block staff","publisher":"The San Francisco Standard","author":"Jacob Clemente","type":"news","date":"2026-02-26"},"references":[{"url":"https://fortune.com/2026/02/27/block-jack-dorsey-ceo-xyz-stock-square-4000-ai-layoffs/","title":"Block CEO Jack Dorsey lays off nearly half of his staff because of AI and predicts most companies will make similar cuts in the next year","publisher":"Fortune","author":"Jake Angelo","type":"news","date":"2026-02-27"}]}
{"id":"2026-02-26-nano-banana-2","date":"2026-02-26","precision":"day","title":"Google releases Nano Banana 2","summary":"Formally Gemini 3.1 Flash Image, it puts the Pro line's world knowledge, text rendering and subject consistency into something fast enough to sit inside a loop. Output goes to 4K.","category":"model","orgs":["Google"],"tier":3,"source":{"url":"https://blog.google/innovation-and-ai/technology/ai/nano-banana-2/","title":"Nano Banana 2: Combining Pro capabilities with lightning-fast speed","publisher":"Google","type":"announcement","date":"2026-02-26"}}
{"id":"2026-02-27-anthropic-supply-chain-designation","date":"2026-02-27","precision":"day","title":"The administration moves to bar Anthropic from federal work","summary":"The President directed agencies to designate Anthropic a supply-chain risk, a label previously reserved for foreign adversaries, following a dispute over the company's usage limits. Anthropic called the designation legally unsound and challenged it in court.","category":"policy","orgs":["Anthropic","United States Department of War"],"tier":2,"source":{"url":"https://www.anthropic.com/news/statement-comments-secretary-war","title":"Statement on the comments from Secretary of War Pete Hegseth","publisher":"Anthropic","type":"announcement","date":"2026-02-27"},"references":[{"url":"https://defensescoop.com/2026/02/27/pentagon-threat-blacklist-anthropic-ai-experts-raise-concerns/","title":"Experts raise questions and concerns about Pentagon's threat to blacklist Anthropic amid AI spat","publisher":"DefenseScoop","type":"news","date":"2026-02-27"},{"url":"https://www.wileyconnect.com/icymi-developments-in-anthropic-challenges-to-department-of-war-supply-chain-risk-designation","title":"ICYMI: Developments in Anthropic Challenges to Department of War Supply Chain Risk Designation","publisher":"Wiley Rein","type":"analysis","date":"2026-04-06"}]}
{"id":"2026-02-28-openai-department-of-war-agreement","date":"2026-02-28","precision":"day","title":"OpenAI signs with the Department of War","summary":"A day after Anthropic was designated a supply chain risk for refusing, OpenAI announced an agreement to run its models on the department's classified network, saying it had secured prohibitions on domestic mass surveillance and human responsibility for the use of force. Those are the two conditions Anthropic would not drop.","category":"deal","orgs":["OpenAI","United States Department of War"],"tier":2,"source":{"url":"https://techcrunch.com/2026/02/28/openais-sam-altman-announces-pentagon-deal-with-technical-safeguards/","title":"OpenAI's Sam Altman announces Pentagon deal with 'technical safeguards'","publisher":"TechCrunch","author":"Anthony Ha","type":"news","date":"2026-02-28"}}
{"id":"2026-03-05-gpt-5-4","date":"2026-03-05","precision":"day","title":"OpenAI releases GPT-5.4","summary":"The coding work from GPT-5.3-Codex pulled back into the general model, aimed at spreadsheets, slide decks and documents. The frontier had started chasing office work.","category":"model","orgs":["OpenAI"],"tier":3,"source":{"url":"https://openai.com/index/introducing-gpt-5-4/","title":"Introducing GPT-5.4","publisher":"OpenAI","type":"announcement","date":"2026-03-05","archiveUrl":"https://web.archive.org/web/20260305182311/https://openai.com/index/introducing-gpt-5-4/"}}
{"id":"2026-03-05-pentagon-anthropic-supply-chain-risk","date":"2026-03-05","precision":"day","title":"The Pentagon designates Anthropic a supply chain risk","summary":"The Department of War made the label official, the first time it has landed on an American company rather than a foreign adversary. Every contractor doing Pentagon work now has to certify that it does not use Anthropic's models.","category":"policy","orgs":["Anthropic","United States Department of War"],"tier":2,"source":{"url":"https://techcrunch.com/2026/03/05/its-official-the-pentagon-has-labeled-anthropic-a-supply-chain-risk/","title":"It's official: The Pentagon has labeled Anthropic a supply-chain risk","publisher":"TechCrunch","author":"Rebecca Bellan","type":"news","date":"2026-03-05"}}
{"id":"2026-03-09-anthropic-sues-pentagon","date":"2026-03-09","precision":"day","title":"Anthropic sues the Pentagon over the supply chain risk label","summary":"Two filings on the same day, one in federal district court in California and one in the DC Circuit, arguing the designation is retaliation for protected speech. Anthropic put the cost at billions of dollars of 2026 revenue and hundreds of millions in private contracts.","category":"policy","orgs":["Anthropic","United States Department of War"],"tier":2,"source":{"url":"https://www.pbs.org/newshour/nation/anthropic-sues-in-federal-court-to-reverse-trump-administrations-supply-chain-risk-designation","title":"Anthropic sues in federal court to reverse Trump administration's 'supply chain risk' designation","publisher":"PBS NewsHour","author":"Matt O'Brien","type":"news","date":"2026-03-09"}}
{"id":"2026-03-09-microsoft-copilot-cowork","date":"2026-03-09","precision":"day","title":"Microsoft ships Copilot Cowork and puts Claude in mainline Copilot","summary":"Anthropic's agent technology arrived inside Microsoft 365 as Copilot Cowork, for work that runs over hours rather than a single turn, and Claude appeared in ordinary Copilot chat alongside OpenAI's models. The company that bankrolled OpenAI now picks a model per task.","category":"product","orgs":["Microsoft","Anthropic"],"tier":2,"source":{"url":"https://www.microsoft.com/en-us/microsoft-365/blog/2026/03/09/powering-frontier-transformation-with-copilot-and-agents/","title":"Powering Frontier Transformation with Copilot and agents","publisher":"Microsoft","author":"Jared Spataro","type":"announcement","date":"2026-03-09"}}
{"id":"2026-03-09-openai-acquires-promptfoo","date":"2026-03-09","precision":"day","title":"OpenAI acquires Promptfoo","summary":"The open source tool that red-teams and evaluates model applications went to OpenAI, to sit inside its Frontier enterprise agent platform with the license unchanged. The software that tests whether agents are safe now belongs to a company that sells agents.","category":"acquisition","orgs":["OpenAI","Promptfoo"],"tier":3,"source":{"url":"https://www.promptfoo.dev/blog/promptfoo-joining-openai/","title":"Promptfoo is joining OpenAI","publisher":"Promptfoo","type":"announcement","date":"2026-03-09"},"references":[{"url":"https://openai.com/index/openai-to-acquire-promptfoo/","title":"OpenAI to acquire Promptfoo","publisher":"OpenAI","type":"announcement","date":"2026-03-09"},{"url":"https://techcrunch.com/2026/03/09/openai-acquires-promptfoo-to-secure-its-ai-agents/","title":"OpenAI acquires Promptfoo to secure its AI agents","publisher":"TechCrunch","type":"news","date":"2026-03-09"}]}
{"id":"2026-03-10-amazon-perplexity-injunction","date":"2026-03-10","precision":"day","title":"A judge blocks Perplexity's shopping agent from Amazon","summary":"Judge Maxine Chesney barred the Comet browser from reaching password-protected parts of Amazon accounts and ordered Perplexity to destroy the data it had already taken. The first court to weigh whether an agent shopping on your behalf counts as you.","category":"policy","orgs":["Amazon","Perplexity"],"tier":2,"source":{"url":"https://www.engadget.com/ai/amazon-wins-a-temporary-injunction-against-perplexitys-comet-browser-184000462.html","title":"Amazon wins a temporary injunction against Perplexity's Comet browser","publisher":"Engadget","author":"Anna Washenko","type":"news","date":"2026-03-10"},"references":[{"url":"https://www.geekwire.com/2026/judge-blocks-perplexitys-ai-bot-from-shopping-on-amazon-in-early-test-of-agentic-commerce/","title":"Judge blocks Perplexity's AI bot from shopping on Amazon in early test of agentic commerce","publisher":"GeekWire","type":"news","date":"2026-03-10"}]}
{"id":"2026-03-10-nvidia-thinking-machines-gigawatt","date":"2026-03-10","precision":"day","title":"NVIDIA commits a gigawatt of Vera Rubin to Thinking Machines Lab","summary":"A multiyear deal to deploy at least a gigawatt of systems that had not shipped yet, plus an investment of undisclosed size in Mira Murati's company. Thirteen months after it was founded, Thinking Machines had compute on the scale only the largest labs have reached.","category":"deal","orgs":["NVIDIA","Thinking Machines Lab"],"tier":2,"source":{"url":"https://thinkingmachines.ai/news/nvidia-partnership/","title":"Thinking Machines Lab and NVIDIA Announce Long-Term Gigawatt-Scale Strategic Partnership","publisher":"Thinking Machines Lab","type":"announcement","date":"2026-03-10"},"references":[{"url":"https://blogs.nvidia.com/blog/nvidia-thinking-machines-lab/","title":"NVIDIA and Thinking Machines Lab Announce Long-Term Gigawatt-Scale Strategic Partnership","publisher":"NVIDIA","type":"announcement","date":"2026-03-10"}]}
{"id":"2026-03-11-china-curbs-openclaw","date":"2026-03-11","precision":"day","title":"China moves to curb OpenClaw at banks and state agencies","summary":"Notices reached government bodies and the largest state banks telling staff not to install the agent on office devices, and to report it for removal if they already had. The country that had spent weeks downloading it decided the permissions it asks for were the problem.","category":"policy","orgs":["OpenClaw"],"tier":2,"source":{"url":"https://www.thestar.com.my/tech/tech-news/2026/03/11/china-moves-to-curb-use-of-openclaw-ai-at-banks-state-agencies","title":"China moves to curb use of OpenClaw AI at banks, state agencies","publisher":"The Star","type":"news","date":"2026-03-11"},"references":[{"url":"https://www.taipeitimes.com/News/front/archives/2026/03/12/2003853672","title":"China scrambles to curb OpenClaw AI at agencies","publisher":"Taipei Times","type":"news","date":"2026-03-12"}]}
{"id":"2026-03-11-google-completes-wiz-acquisition","date":"2026-03-11","precision":"day","title":"Google closes the $32 billion Wiz acquisition","summary":"Cleared and closed almost a year after signing, with Wiz folded into Google Cloud. The biggest cheque Alphabet has ever written went to a security company, not an AI one.","category":"acquisition","orgs":["Google","Wiz"],"tier":2,"source":{"url":"https://www.googlecloudpresscorner.com/2026-03-11-Google-Completes-Acquisition-of-Wiz","title":"Google Completes Acquisition of Wiz","publisher":"Google Cloud","type":"press-release","date":"2026-03-11"},"references":[{"url":"https://techcrunch.com/2026/03/11/google-completes-32b-acquisition-of-wiz/","title":"Google wraps up $32B acquisition of cloud cybersecurity startup Wiz","publisher":"TechCrunch","type":"news","date":"2026-03-11"}]}
{"id":"2026-03-11-replit-series-d","date":"2026-03-11","precision":"day","title":"Replit raises $400 million at $9 billion and ships Agent 4","summary":"Georgian led the round, with Andreessen Horowitz, Coatue and Y Combinator following, and Agent 4 landed the same day with design and code in one surface. The bet is that people who cannot write software will keep paying to have it written.","category":"funding","orgs":["Replit"],"tier":3,"source":{"url":"https://www.prnewswire.com/news-releases/georgian-leads-400m-series-d-investment-in-replit-to-support-continued-investment-in-replit-agent-302711218.html","title":"Georgian Leads $400M Series D Investment in Replit to support continued investment in Replit Agent","publisher":"Replit","type":"press-release","date":"2026-03-11"}}
{"id":"2026-03-16-nvidia-vera-rubin","date":"2026-03-16","precision":"day","title":"NVIDIA launches the Vera Rubin platform","summary":"Seven chips in full production, a rack pairing 72 Rubin GPUs with 36 Vera CPUs, and a claim of ten times Blackwell's inference throughput per watt. Jensen Huang told GTC he saw a trillion dollars of orders through 2027 behind it.","category":"hardware","orgs":["NVIDIA"],"tier":2,"source":{"url":"https://nvidianews.nvidia.com/news/nvidia-vera-rubin-platform","title":"NVIDIA Vera Rubin Opens Agentic AI Frontier","publisher":"NVIDIA","type":"press-release","date":"2026-03-16"},"references":[{"url":"https://www.datacenterknowledge.com/data-center-chips/gtc-2026-nvidia-unveils-vera-rubin-ai-platform-eyes-1t-by-2027","title":"GTC 2026: Nvidia Unveils Vera Rubin AI Platform, Eyes $1T by 2027","publisher":"Data Center Knowledge","author":"Wylie Wong","type":"news","date":"2026-03-17"}]}
{"id":"2026-03-19-cursor-composer-2","date":"2026-03-19","precision":"day","title":"Cursor ships Composer 2","summary":"Anysphere's own coding model at fifty cents per million input tokens, scoring 61.7 on Terminal-Bench 2.0 against Claude Opus 4.6 at 58.0 and GPT-5.4 at 75.1. The third Composer in five months, from a company most people still describe as an editor.","category":"model","orgs":["Anysphere"],"tier":3,"source":{"url":"https://cursor.com/blog/composer-2","title":"Introducing Composer 2","publisher":"Cursor","type":"announcement","date":"2026-03-19"}}
{"id":"2026-03-19-openai-acquires-astral","date":"2026-03-19","precision":"day","title":"OpenAI acquires Astral, the maker of uv and Ruff","summary":"The Python tooling company joined OpenAI to work on Codex, bringing uv, Ruff and ty with it. A large share of the Python ecosystem's build tooling now had one AI lab's name on the maintainer list.","category":"acquisition","orgs":["OpenAI","Astral"],"tier":2,"source":{"url":"https://openai.com/index/openai-to-acquire-astral/","title":"OpenAI to acquire Astral","publisher":"OpenAI","type":"announcement","date":"2026-03-19"},"references":[{"url":"https://astral.sh/blog/openai","title":"Astral to Join OpenAI","publisher":"Astral","type":"announcement","date":"2026-03-19"},{"url":"https://simonwillison.net/2026/Mar/19/openai-acquiring-astral/","title":"Thoughts on OpenAI acquiring Astral and uv/ruff/ty","publisher":"Simon Willison","author":"Simon Willison","type":"analysis","date":"2026-03-19"}]}
{"id":"2026-03-23-claude-computer-use-dispatch","date":"2026-03-23","precision":"day","title":"Claude gets computer use and a phone remote","summary":"Cowork and Claude Code can now point and click at whatever is on the screen when no tool exists for the job, and Dispatch starts that work from a phone so the desk can be empty. Research preview, Pro and Max, macOS and Windows.","category":"product","orgs":["Anthropic"],"tier":2,"source":{"url":"https://claude.com/blog/dispatch-and-computer-use","title":"Put Claude to work on your computer","publisher":"Anthropic","type":"announcement","date":"2026-03-23"},"references":[{"url":"https://www.engadget.com/ai/claude-code-and-cowork-can-now-use-your-computer-210000126.html","title":"Claude Code and Cowork can now use your computer","publisher":"Engadget","author":"Anna Washenko","type":"news","date":"2026-03-23"}]}
{"id":"2026-03-24-sora-shutdown","date":"2026-03-24","precision":"day","title":"OpenAI shuts down Sora","summary":"Less than six months after the launch that topped the App Store, OpenAI said goodbye to the video app in a post on X, redirected the Sora team to world simulation for robotics, and set the API to end in September. The Disney deal ended with it.","category":"product","orgs":["OpenAI"],"tier":2,"source":{"url":"https://www.cnn.com/2026/03/24/tech/openai-sora-video-app-shutting-down","title":"OpenAI is shutting down its Sora video app just months after launch","publisher":"CNN","type":"news","date":"2026-03-24"},"references":[{"url":"https://techcrunch.com/2026/03/24/openais-sora-was-the-creepiest-app-on-your-phone-now-its-shutting-down/","title":"OpenAI's Sora was the creepiest app on your phone. Now it's shutting down","publisher":"TechCrunch","type":"news","date":"2026-03-24"}]}
{"id":"2026-03-25-arc-agi-3","date":"2026-03-25","precision":"day","title":"ARC Prize launches ARC-AGI-3","summary":"Hundreds of hand-built interactive games with no instructions, no stated goal and no rules given, where an agent has to work out what it is doing by playing. Frontier models scored 0.51 percent and humans scored 100.","category":"research","orgs":["ARC Prize"],"tier":2,"source":{"url":"https://arcprize.org/blog/arc-agi-3-launch","title":"Announcing ARC-AGI-3","publisher":"ARC Prize","author":"Greg Kamradt","type":"announcement","date":"2026-03-25"}}
{"id":"2026-03-26-claude-mythos-leak","date":"2026-03-26","precision":"day","title":"A misconfigured cache reveals Claude Mythos","summary":"About 3,000 unpublished files sat reachable on Anthropic's content system, among them a draft post calling Mythos by far the most powerful model the company had built and flagging what it could do to software security. Anthropic confirmed the model exists and called the exposure human error.","category":"culture","orgs":["Anthropic"],"tier":2,"source":{"url":"https://fortune.com/2026/03/26/anthropic-says-testing-mythos-powerful-new-ai-model-after-data-leak-reveals-its-existence-step-change-in-capabilities/","title":"Exclusive: Anthropic acknowledges testing new AI model representing 'step change' in capabilities, after accidental data leak reveals its existence","publisher":"Fortune","author":"Beatrice Nolan","type":"news","date":"2026-03-26"}}
{"id":"2026-03-26-gemini-3-1-flash-live","date":"2026-03-26","precision":"day","title":"Google releases Gemini 3.1 Flash Live","summary":"The audio model behind Gemini Live and Search Live, tuned for lower latency, for hearing pitch and pace, and for holding a conversation in over 200 countries. Everything it says carries a SynthID watermark.","category":"model","orgs":["Google"],"tier":3,"source":{"url":"https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-flash-live/","title":"Gemini 3.1 Flash Live: Making audio AI more natural and reliable","publisher":"Google","type":"announcement","date":"2026-03-26"}}
{"id":"2026-03-27-sam-3-1","date":"2026-03-27","precision":"day","title":"SAM 3.1 tracks sixteen things at once","summary":"Object multiplexing puts up to sixteen tracked objects through a single forward pass, taking a mid-density video from 16 to 32 frames per second on one H100. Checkpoints went to Hugging Face.","category":"model","orgs":["Meta"],"tier":3,"source":{"url":"https://ai.meta.com/blog/segment-anything-model-3/","title":"SAM 3.1: Faster and More Accessible Real-Time Video Detection and Tracking With Multiplexing and Global Reasoning","publisher":"Meta","type":"announcement","date":"2026-03-27"}}
{"id":"2026-03-31-claude-code-source-leak","date":"2026-03-31","precision":"day","title":"Anthropic publishes Claude Code's source by mistake","summary":"Roughly 500,000 lines across 1,900 files went to npm as source rather than compiled output, exposing the harness that tells the model how to use its tools. The second accidental disclosure in five days, and the company again called it human error rather than a breach.","category":"culture","orgs":["Anthropic"],"tier":3,"source":{"url":"https://fortune.com/2026/03/31/anthropic-source-code-claude-code-data-leak-second-security-lapse-days-after-accidentally-revealing-mythos/","title":"Anthropic mistakenly leaks its own AI coding tool's source code, just days after accidentally revealing an upcoming model known as Mythos","publisher":"Fortune","author":"Beatrice Nolan","type":"news","date":"2026-03-31"}}
{"id":"2026-03-31-openai-raises-122-billion","date":"2026-03-31","precision":"day","title":"OpenAI closes a $122 billion round","summary":"The largest private funding round on record, at an $852 billion post-money valuation, with an IPO discussed openly rather than denied. Anthropic beat the record eight weeks later.","category":"funding","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/accelerating-the-next-phase-ai/","title":"OpenAI raises $122 billion to accelerate the next phase of AI","publisher":"OpenAI","type":"announcement","date":"2026-03-31"},"references":[{"url":"https://www.cnbc.com/2026/03/31/openai-funding-round-ipo.html","title":"OpenAI closes record-breaking $122 billion funding round as anticipation builds for IPO","publisher":"CNBC","type":"news","date":"2026-03-31"}]}
{"id":"2026-04-spacex-cursor-option","date":"2026-04","precision":"month","title":"SpaceX buys the right to buy Cursor","summary":"An option to acquire Anysphere for $60 billion in stock or walk away for about $10 billion, alongside a partnership giving Cursor training access to Colossus. SpaceX exercised it eight weeks later, four days after going public.","category":"deal","orgs":["SpaceX","Anysphere"],"tier":2,"source":{"url":"https://finance.yahoo.com/markets/article/spacex-strikes-60-billion-deal-for-the-right-to-buy-ai-coding-startup-cursor-143350832.html","title":"SpaceX strikes $60 billion deal for the right to buy AI coding startup Cursor","publisher":"Yahoo Finance","type":"news","date":"2026-04-23"}}
{"id":"2026-04-02-cursor-3","date":"2026-04-02","precision":"day","title":"Cursor 3 turns the editor into an agent runtime","summary":"The largest redesign since the product shipped in 2023 put an Agents window at the center, for running coding agents in parallel rather than editing files. The IDE stopped being where you type and became where you supervise.","category":"product","orgs":["Anysphere"],"tier":3,"source":{"url":"https://cursor.com/blog/cursor-3","title":"Meet the new Cursor","publisher":"Cursor","type":"announcement","date":"2026-04-02"}}
{"id":"2026-04-06-anthropic-google-broadcom-compute","date":"2026-04-06","precision":"day","title":"Anthropic expands its Google and Broadcom TPU deal","summary":"Anthropic signed an agreement for multiple gigawatts of next-generation Google TPU capacity expected to come online beginning in 2027. Broadcom said about 3.5 gigawatts would be made available through it, subject to Anthropic's continued commercial success.","category":"deal","orgs":["Anthropic","Google","Broadcom"],"tier":2,"source":{"url":"https://www.anthropic.com/news/google-broadcom-partnership-compute","title":"Anthropic expands partnership with Google and Broadcom for multiple gigawatts of next-generation compute","publisher":"Anthropic","type":"announcement","date":"2026-04-06"},"references":[{"url":"https://www.sec.gov/Archives/edgar/data/1730168/000119312526144028/d87999d8k.htm","title":"Broadcom Inc. Form 8-K","publisher":"U.S. Securities and Exchange Commission","type":"report","date":"2026-04-06"},{"url":"https://www.googlecloudpresscorner.com/2026-04-06-Anthropic-Expands-Use-of-Google-Cloud-and-TPUs","title":"Anthropic Expands Use of Google Cloud and TPUs","publisher":"Google Cloud","type":"press-release","date":"2026-04-06"}]}
{"id":"2026-04-07-glm-5-1","date":"2026-04-07","precision":"day","title":"GLM-5.1 puts a downloadable model on top of SWE-Bench Pro","summary":"754 billion parameters under an MIT license, scoring 58.4 and edging past GPT-5.4 and Claude Opus 4.6. The first time an open-weights model led a real-world code repair benchmark rather than trailing it by a year.","category":"model","secondaryCategories":["open-source"],"orgs":["Zhipu AI"],"tier":3,"source":{"url":"https://huggingface.co/zai-org/GLM-5.1","title":"zai-org/GLM-5.1","publisher":"Z.ai","type":"docs","date":"2026-04-07","retrieved":"2026-08-20"},"references":[{"url":"https://www.marktechpost.com/2026/04/08/z-ai-introduces-glm-5-1-an-open-weight-754b-agentic-model-that-achieves-sota-on-swe-bench-pro-and-sustains-8-hour-autonomous-execution/","title":"Z.AI Introduces GLM-5.1: An Open-Weight 754B Agentic Model That Achieves SOTA on SWE-Bench Pro and Sustains 8-Hour Autonomous Execution","publisher":"MarkTechPost","type":"analysis","date":"2026-04-08"}]}
{"id":"2026-04-07-mythos-preview-glasswing","date":"2026-04-07","precision":"day","title":"Anthropic builds a model it will not sell","summary":"Claude Mythos Preview found more than 10,000 high and critical zero-days in its first month, including a seventeen-year-old remote root hole in FreeBSD that it exploited unaided. Anthropic held it back from release and put it behind Project Glasswing, an invitation-only program for critical infrastructure operators.","category":"model","orgs":["Anthropic"],"tier":2,"source":{"url":"https://red.anthropic.com/2026/mythos-preview/","title":"Assessing Claude Mythos Preview's cybersecurity capabilities","publisher":"Anthropic","type":"report","date":"2026-04-07"},"references":[{"url":"https://techcrunch.com/2026/04/07/anthropic-mythos-ai-model-preview-security/","title":"Anthropic debuts preview of powerful new AI model Mythos in new cybersecurity initiative","publisher":"TechCrunch","type":"news","date":"2026-04-07"}]}
{"id":"2026-04-08-anthropic-pentagon-stay-denied","date":"2026-04-08","precision":"day","title":"Anthropic loses its bid to pause the Pentagon blacklist","summary":"The appeals court declined to stay the supply chain risk designation, finding that Anthropic would suffer harm but that the harm looked primarily financial. A California court had enjoined parts of the same action, leaving the company shut out of Defense contracts and free to serve every other agency.","category":"policy","orgs":["Anthropic","United States Department of War"],"tier":3,"source":{"url":"https://www.cnbc.com/2026/04/08/anthropic-pentagon-court-ruling-supply-chain-risk.html","title":"Anthropic loses appeals court bid to temporarily block Pentagon blacklisting","publisher":"CNBC","type":"news","date":"2026-04-08"}}
{"id":"2026-04-08-claude-managed-agents","date":"2026-04-08","precision":"day","title":"Anthropic starts running other people's agents","summary":"Managed Agents moved agent execution onto Anthropic's own infrastructure: isolated containers, default-deny network egress, checkpointing and tracing, billed at token rates plus eight cents per session-hour. The model vendor took the runtime layer.","category":"product","orgs":["Anthropic"],"tier":2,"source":{"url":"https://claude.com/blog/claude-managed-agents","title":"Claude Managed Agents: get to production 10x faster","publisher":"Anthropic","type":"announcement","date":"2026-04-08"},"references":[{"url":"https://www.helpnetsecurity.com/2026/04/09/claude-managed-agents-bring-execution-and-control-to-ai-agent-workflows/","title":"Claude Managed Agents bring execution and control to AI agent workflows","publisher":"Help Net Security","type":"news","date":"2026-04-09"}]}
{"id":"2026-04-08-muse-spark","date":"2026-04-08","precision":"day","title":"Meta returns to the frontier with Muse Spark","summary":"The first model out of Meta Superintelligence Labs after nine months of rebuilding the stack, multimodal, with instant and thinking modes and subagents that run in parallel. It shipped closed, which for the company that published Llama was the news.","category":"model","orgs":["Meta"],"tier":2,"source":{"url":"https://about.fb.com/news/2026/04/introducing-muse-spark-meta-superintelligence-labs/","title":"Introducing Muse Spark: MSL's First Model, Purpose-Built to Prioritize People","publisher":"Meta","type":"announcement","date":"2026-04-08"}}
{"id":"2026-04-10-attacks-on-altman-home","date":"2026-04-10","precision":"day","title":"Two attacks on Sam Altman's home in three days","summary":"A Molotov cocktail thrown at the gate before dawn on Friday, then shots fired from a passing car on Sunday. Arrests followed both, and prosecutors said the first suspect had travelled from Texas intending to kill him.","category":"culture","orgs":["OpenAI"],"tier":2,"source":{"url":"https://www.aljazeera.com/news/2026/4/10/openai-ceo-sam-altmans-home-targeted-in-molotov-cocktail-attack","title":"OpenAI CEO Sam Altman's home targeted in Molotov cocktail attack","publisher":"Al Jazeera","type":"news","date":"2026-04-10"},"references":[{"url":"https://sfstandard.com/2026/04/12/sam-altman-s-home-targeted-second-attack/","title":"Sam Altman's house targeted in second attack; two suspects arrested","publisher":"The San Francisco Standard","type":"news","date":"2026-04-12"},{"url":"https://www.npr.org/2026/04/13/g-s1-117320/openai-sam-altman-molotov-cocktail","title":"Man accused in Molotov cocktail attack of OpenAI CEO's home charged with attempted murder","publisher":"NPR","type":"news","date":"2026-04-13"}]}
{"id":"2026-04-16-claude-opus-4-7","date":"2026-04-16","precision":"day","title":"Claude Opus 4.7 ships, second to a model nobody can buy","summary":"87.6 percent on SWE-bench Verified, a new xhigh effort level, and no price increase. Anthropic said plainly that it does not match Mythos, the model it had declined to release nine days earlier.","category":"model","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/news/claude-opus-4-7","title":"Introducing Claude Opus 4.7","publisher":"Anthropic","type":"announcement","date":"2026-04-16"},"references":[{"url":"https://www.axios.com/2026/04/16/anthropic-claude-opus-model-mythos","title":"Anthropic releases Claude Opus 4.7, concedes it trails unreleased Mythos","publisher":"Axios","type":"news","date":"2026-04-16"}]}
{"id":"2026-04-17-claude-design","date":"2026-04-17","precision":"day","title":"Claude Design turns prompts into decks and prototypes","summary":"A research preview that reads a team's codebase and design files to keep output on brand, exporting to PDF, PPTX and Canva. Aimed at founders and product managers rather than designers, which put Anthropic in Figma and Canva's field of view.","category":"product","orgs":["Anthropic"],"tier":3,"source":{"url":"https://techcrunch.com/2026/04/17/anthropic-launches-claude-design-a-new-product-for-creating-quick-visuals/","title":"Anthropic launches Claude Design, a new product for creating quick visuals","publisher":"TechCrunch","type":"news","date":"2026-04-17"}}
{"id":"2026-04-20-amazon-anthropic-25b","date":"2026-04-20","precision":"day","title":"Amazon adds $25 billion to its Anthropic position","summary":"Five billion immediately at a $380 billion valuation and up to twenty more against commercial milestones, taking Amazon's total to $33 billion. Anthropic committed over $100 billion to AWS across the next decade and locked in up to five gigawatts of capacity.","category":"deal","orgs":["Amazon","Anthropic"],"tier":2,"source":{"url":"https://www.cnbc.com/2026/04/20/amazon-invest-up-to-25-billion-in-anthropic-part-of-ai-infrastructure.html","title":"Amazon to invest up to another $25 billion in Anthropic as part of AI infrastructure deal","publisher":"CNBC","type":"news","date":"2026-04-20"}}
{"id":"2026-04-20-dpa-grid-determination","date":"2026-04-20","precision":"day","title":"Trump invokes the Defense Production Act to expand U.S. grid-equipment production","summary":"An April 20 determination directed the Energy Department to support domestic production of transformers and other grid equipment through purchases and financing. It authorized intervention in constrained supply chains; it did not announce completed capacity or a specific funding award.","category":"policy","orgs":["White House","United States Department of Energy"],"tier":2,"source":{"url":"https://www.whitehouse.gov/presidential-actions/2026/04/presidential-determination-pursuant-to-section-303-of-the-defense-production-act-of-1950-as-amended-on-grid-infrastructure-equipment-and-supply-chain-capacity/","title":"Presidential Determination Pursuant to Section 303 of the Defense Production Act of 1950, as Amended, on Grid Infrastructure, Equipment, and Supply Chain Capacity","publisher":"The White House","type":"docs","date":"2026-04-20"}}
{"id":"2026-04-20-tim-cook-steps-down","date":"2026-04-20","precision":"day","title":"Tim Cook hands Apple to John Ternus","summary":"Cook becomes executive chairman on September 1 after fifteen years as chief executive, and the hardware chief takes the job. Apple's unresolved AI question transfers with it.","category":"company","orgs":["Apple"],"tier":2,"source":{"url":"https://www.apple.com/newsroom/2026/04/tim-cook-to-become-apple-executive-chairman-john-ternus-to-become-apple-ceo/","title":"Tim Cook to become Apple Executive Chairman, John Ternus to become Apple CEO","publisher":"Apple","type":"press-release","date":"2026-04-20"},"references":[{"url":"https://techcrunch.com/2026/04/20/tim-cook-stepping-down-as-apple-ceo-john-ternus-taking-over/","title":"Tim Cook stepping down as Apple CEO, John Ternus taking over","publisher":"TechCrunch","type":"news","date":"2026-04-20"}]}
{"id":"2026-04-21-gpt-image-2","date":"2026-04-21","precision":"day","title":"GPT Image 2 reasons before it draws","summary":"OpenAI put its reasoning stack in front of image generation, with text rendering that holds up in non-Latin scripts, 2K output, and up to eight images that keep the same characters across the set. It reached ChatGPT the next day.","category":"model","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/introducing-chatgpt-images-2-0/","title":"Introducing ChatGPT Images 2.0","publisher":"OpenAI","type":"announcement","date":"2026-04-21"}}
{"id":"2026-04-23-gpt-5-5","date":"2026-04-23","precision":"day","title":"OpenAI releases GPT-5.5","summary":"The fourth point release in eight months, in ChatGPT, Codex and the API on the same day. OpenAI had stopped saving version numbers for occasions.","category":"model","orgs":["OpenAI"],"tier":3,"source":{"url":"https://openai.com/index/introducing-gpt-5-5/","title":"Introducing GPT-5.5","publisher":"OpenAI","type":"announcement","date":"2026-04-23","archiveUrl":"https://web.archive.org/web/20260423180657/https://openai.com/index/introducing-gpt-5-5/"}}
{"id":"2026-04-24-deepseek-v4","date":"2026-04-24","precision":"day","title":"DeepSeek releases V4 preview with MIT-licensed open weights","summary":"DeepSeek released V4-Pro and V4-Flash in preview, with downloadable MIT-licensed weights and a one-million-token context window for each model.","category":"model","secondaryCategories":["open-source"],"orgs":["DeepSeek"],"tier":2,"source":{"url":"https://api-docs.deepseek.com/news/news260424/","title":"DeepSeek V4 Preview Release","publisher":"DeepSeek","type":"announcement","date":"2026-04-24"},"references":[{"url":"https://artificialanalysis.ai/articles/deepseek-is-back-among-the-leading-open-weights-models-with-v4-pro-and-v4-flash","title":"DeepSeek is back among the leading open weights models with V4 Pro and V4 Flash","publisher":"Artificial Analysis","type":"analysis","date":"2026-04-24"}]}
{"id":"2026-04-27-china-blocks-meta-manus","date":"2026-04-27","precision":"day","title":"China orders Meta to unwind its purchase of Manus","summary":"The National Development and Reform Commission prohibited the $2 billion acquisition of the Singapore-based, Chinese-founded agent startup and told the parties to withdraw. Beijing had started policing outbound AI talent the way Washington polices chips.","category":"policy","orgs":["Meta","Manus","National Development and Reform Commission"],"tier":2,"source":{"url":"https://www.cnbc.com/2026/04/27/meta-manus-china-blocks-acquisition-ai-startup.html","title":"China blocks Meta's $2 billion takeover of AI startup Manus","publisher":"CNBC","type":"news","date":"2026-04-27"},"references":[{"url":"https://www.npr.org/2026/04/27/g-s1-118892/china-blocks-meta-from-acquiring-ai-startup-manus","title":"China blocks Meta from acquiring AI startup Manus","publisher":"NPR","type":"news","date":"2026-04-27"}]}
{"id":"2026-04-27-copilot-usage-based-billing","date":"2026-04-27","precision":"day","title":"GitHub Copilot moves to usage-based billing","summary":"From June 1, premium requests give way to a monthly allowance of credits metered on tokens, with subscription prices unchanged. GitHub said it had been absorbing the cost of agents that run for hours on a single question.","category":"product","orgs":["GitHub","Microsoft"],"tier":3,"source":{"url":"https://github.blog/news-insights/company-news/github-copilot-is-moving-to-usage-based-billing/","title":"GitHub Copilot is moving to usage-based billing","publisher":"GitHub","type":"announcement","date":"2026-04-27"}}
{"id":"2026-04-28-openai-models-on-aws","date":"2026-04-28","precision":"day","title":"OpenAI's models reach AWS as Microsoft exclusivity ends","summary":"The cloud exclusivity clause is gone, replaced by an Azure-first commitment, while Microsoft keeps a non-exclusive license through 2032 and its revenue share until 2030 under a cap. OpenAI had outgrown a single cloud.","category":"deal","orgs":["OpenAI","Microsoft","Amazon"],"tier":2,"source":{"url":"https://www.cnbc.com/2026/04/28/openai-brings-models-to-aws-after-ending-exclusivity-with-microsoft.html","title":"OpenAI brings its models to Amazon's cloud after ending exclusivity with Microsoft","publisher":"CNBC","type":"news","date":"2026-04-28"}}
{"id":"2026-04-29-where-the-goblins-came-from","date":"2026-04-29","precision":"day","title":"OpenAI explains the goblins","summary":"Models from GPT-5.1 onward had been working goblins, gremlins and trolls into their metaphors, traced to outsized reward for creature imagery while training the nerdy personality. The fix was an instruction never to mention creatures unless the question calls for one.","category":"research","orgs":["OpenAI"],"tier":3,"source":{"url":"https://openai.com/index/where-the-goblins-came-from/","title":"Where the goblins came from","publisher":"OpenAI","type":"announcement","date":"2026-04-29"}}
{"id":"2026-05-04-anthropic-enterprise-ai-services","date":"2026-05-04","precision":"day","title":"Anthropic and investors form an AI services company","summary":"Anthropic, Blackstone, Hellman & Friedman and Goldman Sachs announced a new AI services company for mid-sized organizations. Its engineers and Anthropic's Applied AI staff would build Claude-powered systems, adding a delivery arm alongside the Claude Partner Network.","category":"deal","orgs":["Anthropic","Blackstone","Hellman & Friedman","Goldman Sachs"],"tier":2,"source":{"url":"https://www.anthropic.com/news/enterprise-ai-services-company","title":"Building a new enterprise AI services company with Blackstone, Hellman & Friedman, and Goldman Sachs","publisher":"Anthropic","type":"announcement","date":"2026-05-04"}}
{"id":"2026-05-05-caisi-frontier-model-agreements","date":"2026-05-05","precision":"day","title":"Google, Microsoft and xAI hand models to the government before release","summary":"The Center for AI Standards and Innovation signed pre-deployment evaluation agreements with three more labs, extending the arrangement OpenAI and Anthropic made in 2024. Five weeks later the same government stopped a model rollout outright.","category":"policy","orgs":["Center for AI Standards and Innovation","Google DeepMind","Microsoft","xAI"],"tier":3,"source":{"url":"https://www.cnbc.com/2026/05/05/ai-oversight-trump-google-microsoft-xai.html","title":"Trump admin moves further into AI oversight, will test Google, Microsoft and xAI models","publisher":"CNBC","type":"news","date":"2026-05-05"},"references":[{"url":"https://www.cnn.com/2026/05/05/tech/microsoft-google-xai-government-test-ai-models","title":"Microsoft, Google and xAI will let the government test their AI models before launch","publisher":"CNN","type":"news","date":"2026-05-05"}]}
{"id":"2026-05-06-anthropic-spacex-colossus","date":"2026-05-06","precision":"day","title":"Anthropic rents Colossus 1, the cluster built to train Grok","summary":"Over 300 megawatts and more than 220,000 Nvidia GPUs inside a month, with the first of it spent on doubling Claude Code rate limits. Reporting two weeks later put the price at $1.25 billion a month through 2029.","category":"deal","orgs":["Anthropic","SpaceX","xAI"],"tier":2,"source":{"url":"https://www.anthropic.com/news/higher-limits-spacex","title":"Higher usage limits for Claude and a compute deal with SpaceX","publisher":"Anthropic","type":"announcement","date":"2026-05-06"},"references":[{"url":"https://x.ai/news/anthropic-compute-partnership","title":"New Compute Partnership with Anthropic","publisher":"SpaceXAI","type":"announcement","date":"2026-05-06"},{"url":"https://techcrunch.com/2026/05/20/anthropic-will-pay-xai-1-25-billion-per-month-for-compute/","title":"Anthropic will pay xAI $1.25B per month for compute","publisher":"TechCrunch","type":"news","date":"2026-05-20"}]}
{"id":"2026-05-06-claude-agents-dreaming","date":"2026-05-06","precision":"day","title":"Claude agents get to dream","summary":"Dreaming runs between sessions, replays what an agent did and writes what it learned into memory, which Anthropic compares to the consolidation a brain does in sleep. It shipped at Code with Claude alongside a grader that sends work back for revision and an orchestrator that fans jobs out to subagents.","category":"product","orgs":["Anthropic"],"tier":3,"source":{"url":"https://thenewstack.io/anthropic-managed-agents-dreaming-outcomes/","title":"Anthropic will let its managed agents dream","publisher":"The New Stack","type":"news","date":"2026-05-06"},"references":[{"url":"https://claude.com/blog/new-in-claude-managed-agents","title":"New in Claude Managed Agents: dreaming, outcomes, and multiagent orchestration","publisher":"Anthropic","type":"announcement","date":"2026-05-19"},{"url":"https://claude.com/blog/code-w-claude-sf-2026-sf","title":"Code w/ Claude SF 2026: Building on the AI exponential","publisher":"Anthropic","type":"announcement","date":"2026-05-12"}]}
{"id":"2026-05-11-openai-deployment-company","date":"2026-05-11","precision":"day","title":"OpenAI starts a consulting firm","summary":"The OpenAI Deployment Company launched with more than $4 billion from nineteen investors led by TPG, and bought the consultancy Tomoro for its 150 forward deployed engineers. The lab that sold an API now sells the people who install it.","category":"company","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/openai-launches-the-deployment-company/","title":"OpenAI launches the OpenAI Deployment Company to help businesses build around intelligence","publisher":"OpenAI","type":"announcement","date":"2026-05-11"},"references":[{"url":"https://www.axios.com/2026/05/11/openai-deployco-private-equity","title":"OpenAI launches AI consulting arm valued at $14 billion","publisher":"Axios","type":"news","date":"2026-05-11"}]}
{"id":"2026-05-11-thinking-machines-interaction-models","date":"2026-05-11","precision":"day","title":"Thinking Machines builds a model that listens while it talks","summary":"TML-Interaction-Small drops turn-taking entirely, cutting audio, video and text into 200 millisecond micro-turns so the model can interrupt and be interrupted. Mira Murati's lab called it a model category rather than a feature, and put it behind a research preview.","category":"model","orgs":["Thinking Machines Lab"],"tier":2,"source":{"url":"https://thinkingmachines.ai/blog/interaction-models/","title":"Interaction Models: A Scalable Approach to Human-AI Collaboration","publisher":"Thinking Machines Lab","type":"announcement","date":"2026-05-11"},"references":[{"url":"https://techcrunch.com/2026/05/11/thinking-machines-wants-to-build-an-ai-that-actually-listens-while-it-talks/","title":"Thinking Machines wants to build an AI that actually listens while it talks","publisher":"TechCrunch","type":"news","date":"2026-05-11"}]}
{"id":"2026-05-12-claude-for-legal","date":"2026-05-12","precision":"day","title":"Anthropic ships Claude for Legal","summary":"Twelve practice-area plugins and more than twenty connectors into Westlaw, Docusign and the rest of the legal stack, including one for Harvey, a legal AI company built on Claude. The model vendor moved into its customer's market and the customer wired itself in anyway.","category":"product","orgs":["Anthropic","Harvey"],"tier":3,"source":{"url":"https://techcrunch.com/2026/05/12/the-ai-legal-services-industry-is-heating-up-anthropic-is-getting-in-on-the-action/","title":"The AI legal services industry is heating up — Anthropic is getting in on the action","publisher":"TechCrunch","type":"news","date":"2026-05-12"},"references":[{"url":"https://www.lawnext.com/2026/05/anthropic-goes-all-in-on-legal-releasing-more-than-20-connectors-and-12-practice-area-plugins-for-claude.html","title":"Anthropic Goes All-In on Legal, Releasing More Than 20 Connectors and 12 Practice-Area Plugins for Claude","publisher":"LawSites","type":"analysis","date":"2026-05-12"}]}
{"id":"2026-05-14-cerebras-ipo","date":"2026-05-14","precision":"day","title":"Cerebras goes public and doubles on day one","summary":"Priced at $185 above an already raised range, opened at $385 and closed up 68 percent, raising $5.55 billion at a $95 billion market cap. The largest US tech listing since Uber in 2019, for a company whose product is a single wafer-sized chip.","category":"company","orgs":["Cerebras"],"tier":2,"source":{"url":"https://www.cnbc.com/2026/05/14/cerebras-cbrs-stock-trade-nasdaq-ipo.html","title":"Cerebras pops 68% in Nasdaq debut, pushing the AI chipmaker's market cap to $95 billion","publisher":"CNBC","type":"news","date":"2026-05-14"},"references":[{"url":"https://techcrunch.com/2026/05/14/cerebras-raises-5-5b-kicking-off-2026s-ipo-season-with-a-bang/","title":"Cerebras raises $5.5B, then stock pops 108%, in the first huge tech IPO of 2026","publisher":"TechCrunch","type":"news","date":"2026-05-14"}]}
{"id":"2026-05-14-codex-chatgpt-mobile","date":"2026-05-14","precision":"day","title":"Codex arrives in the ChatGPT mobile app","summary":"Approve a diff, redirect a running task or read terminal output from a phone, on every plan including the free tier. It only talks to the Mac app, so the agent still needs a desk somewhere.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://techcrunch.com/2026/05/14/openai-says-codex-is-coming-to-your-phone/","title":"OpenAI says Codex is coming to your phone","publisher":"TechCrunch","type":"news","date":"2026-05-14"}}
{"id":"2026-05-18-cloudflare-project-glasswing","date":"2026-05-18","precision":"day","title":"Cloudflare points Mythos at its own code","summary":"Across more than fifty internal repositories the model chained low-severity bugs into working exploits and wrote proofs that compiled, where earlier models stopped at describing the bug and left exploitability unknown. Cloudflare's conclusion was that this is a different kind of tool, not a better one.","category":"research","orgs":["Cloudflare","Anthropic"],"tier":2,"source":{"url":"https://blog.cloudflare.com/cyber-frontier-models/","title":"Project Glasswing: what Mythos showed us","publisher":"Cloudflare","type":"report","date":"2026-05-18"}}
{"id":"2026-05-18-cursor-composer-2-5","date":"2026-05-18","precision":"day","title":"Composer 2.5 matches the frontier at a tenth the price","summary":"Cursor's own model scored 79.8 percent on SWE-Bench Multilingual against Claude Opus 4.7 and GPT-5.5, at 50 cents per million input tokens. Two months after Composer 2, in the middle of a compute shortage that made cheap tokens the whole argument.","category":"model","orgs":["Anysphere"],"tier":3,"source":{"url":"https://cursor.com/blog/composer-2-5","title":"Introducing Composer 2.5","publisher":"Cursor","type":"announcement","date":"2026-05-18"}}
{"id":"2026-05-18-musk-loses-openai-case","date":"2026-05-18","precision":"day","title":"Musk loses his case against OpenAI","summary":"The suit claiming OpenAI had betrayed its founding nonprofit mission ended in a loss, filed in 2024 and decided seven months after the restructuring it was meant to stop.","category":"company","orgs":["OpenAI","xAI"],"tier":3,"source":{"url":"https://techcrunch.com/2026/05/18/elon-musk-has-lost-his-lawsuit-against-sam-altman-and-openai/","title":"Elon Musk has lost his lawsuit against Sam Altman and OpenAI","publisher":"TechCrunch","type":"news","date":"2026-05-18"}}
{"id":"2026-05-19-gemini-3-5","date":"2026-05-19","precision":"day","title":"Gemini 3.5 leads with the small model","summary":"Google shipped the Flash tier first and claimed it beat Gemini 3.1 Pro on coding and agentic work while running four times faster, at 76.2 percent on Terminal-Bench 2.1. Pro was promised for the following month, which is a new order of operations for a frontier release.","category":"model","orgs":["Google DeepMind"],"tier":2,"source":{"url":"https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/","title":"Gemini 3.5: frontier intelligence with action","publisher":"Google","type":"announcement","date":"2026-05-19"}}
{"id":"2026-05-19-gemini-omni","date":"2026-05-19","precision":"day","title":"Gemini Omni generates video and calls it a world model","summary":"Any input to any output, starting with video, with edits that build on one another while physics and characters hold across them. Google framed it as the move from predicting text to simulating reality, which is a large claim for something that launched inside YouTube Shorts.","category":"model","orgs":["Google DeepMind"],"tier":2,"source":{"url":"https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-omni/","title":"Introducing Gemini Omni","publisher":"Google","type":"announcement","date":"2026-05-19"}}
{"id":"2026-05-19-gemini-spark","date":"2026-05-19","precision":"day","title":"Gemini Spark runs on Google's machines, not yours","summary":"A personal agent on a dedicated cloud VM, built on the Antigravity harness, working through Gmail, Docs and Slides with the laptop closed. Google's answer to the coding agents, aimed at everyone who does not write code.","category":"product","orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://blog.google/innovation-and-ai/products/gemini-app/next-evolution-gemini-app/","title":"The Gemini app becomes more agentic, delivering proactive, 24/7 help","publisher":"Google","type":"announcement","date":"2026-05-19"}}
{"id":"2026-05-19-karpathy-joins-anthropic","date":"2026-05-19","precision":"day","title":"Andrej Karpathy joins Anthropic","summary":"The OpenAI founding member and former Tesla AI director, who had spent three years teaching the field from the outside, went back inside at a competitor. Six months earlier he had said agents would take a decade to work through their problems.","category":"company","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.axios.com/2026/05/19/anthropic-openai-karpathy-andrej-claude","title":"OpenAI co-founder Andrej Karpathy joins Anthropic","publisher":"Axios","type":"news","date":"2026-05-19"},"references":[{"url":"https://www.dwarkesh.com/p/andrej-karpathy","title":"Andrej Karpathy: It will take a decade to work through the issues with agents","publisher":"Dwarkesh Podcast","author":"Dwarkesh Patel","type":"video","date":"2025-10-17"}]}
{"id":"2026-05-19-openai-guaranteed-capacity","date":"2026-05-19","precision":"day","title":"OpenAI starts selling compute the way clouds sell it","summary":"One to three year commitments with discounts that deepen by term, up to a billion tokens a minute, drawn down across products. Inference had become scarce enough to reserve in advance.","category":"deal","orgs":["OpenAI"],"tier":3,"source":{"url":"https://openai.com/business/guaranteed-capacity/","title":"OpenAI Guaranteed Capacity","publisher":"OpenAI","type":"announcement","date":"2026-05-19"},"references":[{"url":"https://www.cnbc.com/2026/05/19/openai-announces-new-guaranteed-capacity-offering-for-customers-to-secure-compute.html","title":"OpenAI announces new Guaranteed Capacity offering for customers to secure compute","publisher":"CNBC","type":"news","date":"2026-05-19"}]}
{"id":"2026-05-20-openai-yc-tokens-for-equity","date":"2026-05-20","precision":"day","title":"OpenAI offers every YC startup $2 million in tokens for equity","summary":"An uncapped SAFE paid in API credits rather than cash, across a batch of roughly 169 companies. The lab had found a way to buy equity with the thing it makes.","category":"deal","orgs":["OpenAI","Y Combinator"],"tier":3,"source":{"url":"https://techcrunch.com/2026/05/20/sam-altman-makes-mic-drop-offer-to-every-y-combinator-startup/","title":"Sam Altman makes 'mic drop' offer to every Y Combinator startup","publisher":"TechCrunch","type":"news","date":"2026-05-20"}}
{"id":"2026-05-20-unit-distance-conjecture","date":"2026-05-20","precision":"day","title":"An OpenAI model disproves an eighty-year-old conjecture","summary":"A counterexample to the unit distance conjecture in discrete geometry, a problem Erdős posed in the 1940s and nobody had closed since. The week that followed was spent arguing over how much of the work was the model's.","category":"research","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/model-disproves-discrete-geometry-conjecture/","title":"An OpenAI model has disproved a central conjecture in discrete geometry","publisher":"OpenAI","type":"announcement","date":"2026-05-20"},"references":[{"url":"https://garymarcus.substack.com/p/checking-the-math-behind-openai-and","title":"Checking the math behind OpenAI and Anthropic's latest headlines","publisher":"Marcus on AI","author":"Gary Marcus","type":"analysis","date":"2026-05-21"},{"url":"https://news.ycombinator.com/item?id=48212493","title":"An OpenAI model has disproved a central conjecture in discrete geometry","publisher":"Hacker News","type":"analysis","date":"2026-05-21"}]}
{"id":"2026-05-23-deepseek-permanent-price-cut","date":"2026-05-23","precision":"day","title":"DeepSeek makes its 75 percent discount permanent","summary":"DeepSeek made the V4-Pro promotional discount, previously due to expire May 31, permanent. Its published rates ranged from $0.003625 to $0.87 per million tokens, depending on usage type.","category":"product","orgs":["DeepSeek"],"tier":3,"source":{"url":"https://www.engadget.com/2180062/deepseek-permanently-reduces-the-price-of-its-flagship-v4-model-by-75-percent/","title":"DeepSeek permanently reduces the price of its flagship V4 model by 75 percent","publisher":"Engadget","type":"news","date":"2026-05-23"}}
{"id":"2026-05-25-magnifica-humanitas","date":"2026-05-25","precision":"day","title":"Pope Leo XIV makes AI the subject of his first encyclical","summary":"Magnifica humanitas argues that artificial intelligence must be disarmed, and that human worth is not a capability score. He signed it on the 135th anniversary of Rerum novarum, the encyclical that set the church's position on industrial labor, and presented it himself.","category":"culture","orgs":["Holy See"],"tier":2,"source":{"url":"https://www.vaticannews.va/en/pope/news/2026-05/pope-leo-xiv-encyclical-magnifica-humanitas-ai.html","title":"Pope Leo's 'Magnifica humanitas': AI must serve humanity not concentrate power","publisher":"Vatican News","type":"announcement","date":"2026-05-25"},"references":[{"url":"https://www.vatican.va/content/leo-xiv/en/encyclicals/documents/20260515-magnifica-humanitas.html","title":"Encyclical Letter of His Holiness Leo XIV Magnifica Humanitas","publisher":"Holy See","type":"docs","date":"2026-05-15","locator":"Dated 15 May 2026, published 25 May 2026","retrieved":"2026-08-20"},{"url":"https://www.usccb.org/news/2026/first-encyclical-pope-leo-urges-world-disarm-ai-amid-increased-reliance","title":"In first encyclical, Pope Leo urges world to 'disarm' AI amid increased reliance","publisher":"United States Conference of Catholic Bishops","type":"news","date":"2026-05-25"}]}
{"id":"2026-05-26-openrouter-series-b","date":"2026-05-26","precision":"day","title":"OpenRouter raises $113 million as routing becomes infrastructure","summary":"CapitalG led at a $1.3 billion valuation, with the venture arms of Nvidia, Databricks, Snowflake and MongoDB alongside. The company was moving 25 trillion tokens a week across more than 400 models and taking about five percent of the spend.","category":"funding","orgs":["OpenRouter"],"tier":3,"source":{"url":"https://techcrunch.com/2026/05/26/openrouter-more-than-doubles-valuation-to-1-3b-in-a-year/","title":"OpenRouter more than doubles valuation to $1.3B in a year","publisher":"TechCrunch","type":"news","date":"2026-05-26"},"references":[{"url":"https://siliconangle.com/2026/05/26/openrouter-raises-113m-bring-order-enterprise-ai-inference-routing/","title":"OpenRouter raises $113M to bring order to enterprise AI inference routing","publisher":"SiliconANGLE","type":"news","date":"2026-05-26"}]}
{"id":"2026-05-27-cognition-raises-1b","date":"2026-05-27","precision":"day","title":"Cognition raises $1 billion at $26 billion","summary":"Two and a half times its September valuation, on $492 million of annualized revenue and enterprise usage growing 50 percent a month. Devin was writing 89 percent of the production code at the company that sells Devin.","category":"funding","orgs":["Cognition"],"tier":3,"source":{"url":"https://techcrunch.com/2026/05/27/ai-coding-startup-cognition-raises-1b-at-25b-pre-money-valuation/","title":"AI coding startup Cognition raises $1B at $25B pre-money valuation","publisher":"TechCrunch","type":"news","date":"2026-05-27"}}
{"id":"2026-05-28-anthropic-series-h","date":"2026-05-28","precision":"day","title":"Anthropic raises $65 billion at $965 billion","summary":"The largest private round ever raised, closing eight weeks after OpenAI set the previous record and passing it on valuation. Anthropic shipped Claude Opus 4.8 the same day, which is one way to justify the number.","category":"funding","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/news/series-h","title":"Anthropic raises $65B in Series H funding at $965B post-money valuation","publisher":"Anthropic","type":"announcement","date":"2026-05-28"},"references":[{"url":"https://www.nytimes.com/2026/05/12/technology/anthropic-funding-950-billion-valuation.html","title":"Anthropic in Talks to Raise Funding at a $950 Billion Valuation","publisher":"The New York Times","type":"news","date":"2026-05-12"},{"url":"https://www.anthropic.com/news/claude-opus-4-8","title":"Introducing Claude Opus 4.8","publisher":"Anthropic","type":"announcement","date":"2026-05-28"}]}
{"id":"2026-05-28-claude-opus-4-8","date":"2026-05-28","precision":"day","title":"Claude Opus 4.8","summary":"Shipped the same day as the $65 billion round, and presented as a refinement rather than a jump. It was the most capable model Anthropic would sell anyone at the time, with the Mythos class still in preview.","category":"model","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/claude-opus-4-8","title":"Introducing Claude Opus 4.8","publisher":"Anthropic","type":"announcement","date":"2026-05-28"}}
{"id":"2026-06-02-executive-order-14409","date":"2026-06-02","precision":"day","title":"The White House orders a voluntary framework for frontier-model cyber review","summary":"Executive Order 14409 ordered agencies to issue cyber-defense directives and develop a classified model-cyber benchmark. It also directed them to design a voluntary pre-release framework for \"covered frontier models,\" while expressly disclaiming any mandatory licensing, preclearance or permitting system.","category":"policy","orgs":["White House"],"tier":2,"source":{"url":"https://www.govinfo.gov/content/pkg/DCPD-202600376/html/DCPD-202600376.htm","title":"Executive Order 14409—Promoting Advanced Artificial Intelligence Innovation and Security","publisher":"Office of the Federal Register, National Archives and Records Administration","type":"docs","date":"2026-06-02"}}
{"id":"2026-06-02-microsoft-frontier-tuning","date":"2026-06-02","precision":"day","title":"Microsoft previews Frontier Tuning for enterprise AI agents","summary":"A private preview shown at Build 2026: reinforcement learning applied inside a customer's own compliance boundary, on their own data and workflows, to produce tuned models and a runtime harness. Microsoft's bet is that the most valuable AI in an organization is the one that already knows how it works.","category":"product","orgs":["Microsoft"],"tier":3,"source":{"url":"https://devblogs.microsoft.com/microsoft365dev/frontier-tuning-teaching-ai-to-work-the-way-you-do/","title":"Frontier Tuning: Teaching AI to work the way you do","publisher":"Microsoft","type":"announcement","date":"2026-06-02"}}
{"id":"2026-06-03-gemma-4","date":"2026-06-03","precision":"day","title":"Gemma 4 drops the vision encoder","summary":"Google's open-weights model went encoder-free and multimodal in one architecture, at a size that runs on a single consumer card. Open weights from US labs had become a rarity worth noting.","category":"model","secondaryCategories":["open-source"],"orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12b/","title":"Gemma 4 12B: A unified, encoder-free multimodal model","publisher":"Google","type":"announcement","date":"2026-06-03"}}
{"id":"2026-06-03-qwen-3-7-plus","date":"2026-06-03","precision":"day","title":"Qwen3.7-Plus adds vision to Alibaba's agent model, drops open weights","summary":"A multimodal upgrade to Alibaba's agent line, pairing vision and video understanding with a million-token context window at roughly a sixth of the flagship's price. Unlike its predecessors, it shipped API-only, breaking Qwen's open-weight habit.","category":"model","orgs":["Alibaba"],"tier":3,"source":{"url":"https://www.alibabacloud.com/blog/qwen3-7-plus-multimodal-agent-intelligence_603206","title":"Qwen3.7-Plus: Multimodal Agent Intelligence","publisher":"Alibaba Cloud","type":"announcement","date":"2026-06-03"}}
{"id":"2026-06-07-nvidia-sk-hynix-memory-partnership","date":"2026-06-07","precision":"day","title":"Nvidia and SK hynix sign a multiyear deal on next-generation AI memory","summary":"A technology partnership to codevelop memory for Nvidia's Vera Rubin systems, with SK hynix in turn using Nvidia's tools to accelerate its own chip design and fab operations. Memory, not compute, had become the AI buildout's tightest bottleneck.","category":"deal","orgs":["NVIDIA","SK hynix"],"tier":3,"source":{"url":"https://nvidianews.nvidia.com/news/sk-hynix-ai-factory","title":"NVIDIA and SK hynix Announce Multiyear Technology Partnership to Advance Memory for AI Factories","publisher":"NVIDIA","type":"press-release","date":"2026-06-07"}}
{"id":"2026-06-08-openai-confidential-ipo-filing","date":"2026-06-08","precision":"day","title":"OpenAI confidentially files for an IPO","summary":"A draft registration filed with the SEC, unpriced and undated, just over a week after Anthropic filed to do the same. OpenAI said it had not decided on timing, since some of what it wants to do is still easier done as a private company.","category":"company","orgs":["OpenAI"],"tier":2,"source":{"url":"https://techcrunch.com/2026/06/08/following-anthropic-openai-files-confidentially-for-ipo/","title":"OpenAI files confidentially for IPO, following Anthropic","publisher":"TechCrunch","type":"news","date":"2026-06-08"},"references":[{"url":"https://openai.com/index/openai-submits-confidential-s-1/","title":"Confidential submission of draft S-1 to the SEC","publisher":"OpenAI","type":"announcement","date":"2026-06-08"}]}
{"id":"2026-06-09-claude-fable-5-mythos-5","date":"2026-06-09","precision":"day","title":"Anthropic ships Mythos 5 and a safer twin","summary":"Two models on one day: Mythos 5 at the top of Anthropic's capability range, and Fable 5, described as a Mythos-class model made safe for general release. Three days later the United States government suspended access to both.","category":"model","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/news/claude-fable-5-mythos-5","title":"Claude Fable 5 and Claude Mythos 5","publisher":"Anthropic","type":"announcement","date":"2026-06-09"},"references":[{"url":"https://aws.amazon.com/blogs/aws/anthropic-claude-fable-5-on-aws-mythos-class-capabilities-with-built-in-safeguards-now-available/","title":"Anthropic Claude Fable 5 on AWS: Mythos-class capabilities with built-in safeguards now available","publisher":"Amazon Web Services","author":"Channy Yun","type":"announcement","date":"2026-06-09"}]}
{"id":"2026-06-11-prometheus-series-b","date":"2026-06-11","precision":"day","title":"Prometheus closes a $12 billion Series B","summary":"Jeff Bezos said Prometheus had closed a $12 billion Series B, funding compute and dataset creation for an AI system it describes as an \"artificial general engineer.\" The direct account is an interview, not a company financing release or an audited disclosure.","category":"funding","orgs":["Prometheus"],"tier":2,"source":{"url":"https://pressroom.versantmedia.com/cnbc/press-releases/cnbc-exclusive-transcript-prometheus-co-founders-and-co-ceos-jeff-bezos-and-vik","title":"CNBC Exclusive Transcript: Prometheus Co-Founders and Co-CEOs Jeff Bezos and Vik Bajaj","publisher":"CNBC","type":"news","date":"2026-06-11","locator":"Bezos says, \"we just concluded our Series B\" and later refers to \"just closing a $12 billion round.\""},"references":[{"url":"https://techcrunch.com/2026/06/11/jeff-bezoss-prometheus-raises-12b-to-build-an-artificial-general-engineer-for-the-physical-world/","title":"Jeff Bezos's Prometheus raises $12B to build an 'artificial general engineer' for the physical world","publisher":"TechCrunch","author":"Marina Temkin","type":"news","date":"2026-06-11"}]}
{"id":"2026-06-12-frontier-model-export-controls","date":"2026-06-12","precision":"day","title":"The United States puts export controls on frontier models","summary":"A Commerce directive suspended all foreign-national access to Anthropic's Fable 5 and Mythos 5, and held back OpenAI's GPT-5.6 rollout on the same grounds. Access to a frontier model had become a controlled export.","category":"policy","orgs":["Anthropic","OpenAI","United States Department of Commerce"],"tier":1,"source":{"url":"https://www.anthropic.com/news/fable-mythos-access","title":"Statement on the US government directive to suspend access to Fable 5 and Mythos 5","publisher":"Anthropic","type":"announcement","date":"2026-06-12"},"references":[{"url":"https://www.csis.org/analysis/department-commerce-restricted-access-anthropics-latest-models-what-comes-next","title":"The Department of Commerce Restricted Access to Anthropic's Latest Models. What Comes Next?","publisher":"Center for Strategic and International Studies","type":"analysis","date":"2026-06-16"},{"url":"https://www.anthropic.com/news/redeploying-fable-5","title":"Redeploying Claude Fable 5","publisher":"Anthropic","type":"announcement","date":"2026-06-30"}]}
{"id":"2026-06-12-kimi-k2-7-code","date":"2026-06-12","precision":"day","title":"Moonshot ships Kimi K2.7 Code, a trillion-parameter open coding model","summary":"An open-weight upgrade to Moonshot's coding line, with double-digit gains on its own coding benchmarks and 30 percent fewer reasoning tokens than K2.6. Released on Hugging Face under a modified MIT license, one more entry in China's open-weights coding wave.","category":"model","secondaryCategories":["open-source"],"orgs":["Moonshot AI"],"tier":3,"source":{"url":"https://devops.com/moonshot-ais-kimi-k2-7-code-targets-token-efficiency-in-agentic-coding/","title":"Moonshot AI's Kimi K2.7-Code Targets Token Efficiency in Agentic Coding","publisher":"DevOps.com","type":"news","date":"2026-06-15"}}
{"id":"2026-06-12-spacex-ipo","date":"2026-06-12","precision":"day","title":"SpaceX goes public in the largest IPO ever","summary":"About $75 billion raised on the Nasdaq, roughly triple the previous record, with the stock closing up 19 percent. Four days later the company spent that paper on Cursor.","category":"company","orgs":["SpaceX"],"tier":2,"source":{"url":"https://www.nbcnews.com/business/markets/spacex-ipo-stock-price-rcna349760","title":"SpaceX stock gains 19% its first trading day, closing out a historic IPO","publisher":"NBC News","type":"news","date":"2026-06-12"},"references":[{"url":"https://www.cnbc.com/2026/06/11/spacex-raises-75-billion-in-record-setting-ipo-ahead-of-nasdaq-debut.html","title":"SpaceX raising $75 billion in record-setting IPO as Nasdaq debut awaits","publisher":"CNBC","type":"news","date":"2026-06-11"}]}
{"id":"2026-06-16-glm-5-2-leads-open-weights","date":"2026-06-16","precision":"day","title":"GLM-5.2 takes the open-weights lead","summary":"Zhipu's model topped the independent open-weights index, and a security vendor's own benchmarks put it ahead of Claude on cyber tasks two weeks later. The best model you could download was Chinese again.","category":"model","secondaryCategories":["open-source"],"orgs":["Zhipu AI"],"tier":3,"source":{"url":"https://artificialanalysis.ai/articles/glm-5-2-is-the-new-leading-open-weights-model-on-the-artificial-analysis-intelligence-index","title":"GLM-5.2 is the new leading open weights model on the Artificial Analysis Intelligence Index","publisher":"Artificial Analysis","type":"analysis","date":"2026-06-16"},"references":[{"url":"https://semgrep.dev/blog/2026/we-have-mythos-at-home-glm-52-beats-claude-in-our-cyber-benchmarks/","title":"GLM 5.2 beats Claude in our cyber benchmarks","publisher":"Semgrep","type":"analysis","date":"2026-06-28"}]}
{"id":"2026-06-16-spacex-acquires-cursor","date":"2026-06-16","precision":"day","title":"SpaceX buys Cursor for $60 billion","summary":"The largest acquisition of a venture-backed startup on record, all stock, for a code editor that did not exist three years earlier. The tools for writing software had become worth more than most of the software.","category":"acquisition","orgs":["SpaceX","Anysphere"],"tier":1,"source":{"url":"https://www.cnbc.com/2026/06/16/spacex-spcx-cursor-acquisition-ipo.html","title":"SpaceX to acquire the AI coding startup Cursor for $60 billion","publisher":"CNBC","type":"news","date":"2026-06-16"},"references":[{"url":"https://www.forbes.com/sites/sandycarter/2026/06/16/spacex-buys-cursor-in-largest-startup-acquisition-ever-at-60-billion/","title":"SpaceX Buys Cursor In Largest Startup Acquisition Ever At $60 Billion","publisher":"Forbes","author":"Sandy Carter","type":"news","date":"2026-06-16"}]}
{"id":"2026-06-20-john-jumper-leaves-deepmind-for-anthropic","date":"2026-06-20","precision":"day","title":"Nobel laureate John Jumper leaves Google DeepMind for Anthropic","summary":"AlphaFold's co-creator departed after nearly nine years, one of several senior researchers to leave DeepMind for a rival that month. Anthropic had spent the year building out wet labs and an AI-for-science push; Jumper's arrival explained why.","category":"culture","orgs":["Google DeepMind","Anthropic"],"tier":2,"source":{"url":"https://techcrunch.com/2026/06/20/nobel-laureate-john-jumper-is-leaving-deepmind-for-rival-anthropic/","title":"Nobel laureate John Jumper is leaving DeepMind for rival Anthropic","publisher":"TechCrunch","type":"news","date":"2026-06-20"},"references":[{"url":"https://www.cnbc.com/2026/06/19/john-jumper-to-leave-google-deepmind-for-anthropic.html","title":"John Jumper to leave Google DeepMind for Anthropic","publisher":"CNBC","type":"news","date":"2026-06-19"}]}
{"id":"2026-06-23-claude-tag","date":"2026-06-23","precision":"day","title":"Anthropic puts Claude in Slack as a shared teammate, not a chatbot","summary":"Claude Tag lets anyone in a channel tag @Claude in and delegate a task, building context across the whole conversation instead of one message at a time. It replaces Anthropic's earlier Slack app; the company says 65 percent of its own product team's code now comes from an internal version.","category":"product","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/introducing-claude-tag","title":"Introducing Claude Tag","publisher":"Anthropic","type":"announcement","date":"2026-06-23"},"references":[{"url":"https://techcrunch.com/2026/06/23/anthropics-claude-tag-is-learning-your-company-one-slack-message-at-a-time/","title":"Anthropic's Claude Tag is learning your company, one Slack message at a time","publisher":"TechCrunch","type":"news","date":"2026-06-23"}]}
{"id":"2026-06-24-anthropic-alibaba-distillation-accusation","date":"2026-06-24","precision":"day","title":"Anthropic accuses Alibaba of its largest distillation attack yet","summary":"A letter to the Senate Banking Committee alleged 28.8 million exchanges with Claude through 25,000 fraudulent accounts between April and June, aimed at reproducing Claude's coding and agentic capabilities in Alibaba's own Qwen models. Alibaba did not respond publicly.","category":"company","orgs":["Anthropic","Alibaba"],"tier":2,"source":{"url":"https://www.cnbc.com/2026/06/24/anthropic-alibaba-distillation-campaign.html","title":"Anthropic accuses Alibaba of campaign to 'brazenly' and 'illicitly' extract AI capabilities","publisher":"CNBC","type":"news","date":"2026-06-24"}}
{"id":"2026-06-24-openai-broadcom-jalapeno-chip","date":"2026-06-24","precision":"day","title":"OpenAI unveils Jalapeño, its first chip, built with Broadcom","summary":"First samples of a custom inference processor tuned for OpenAI's own workloads, aimed first at coding. Broadcom's CEO put early cost savings at roughly 50 percent against standard GPUs, with initial deployment targeted for late 2026.","category":"hardware","orgs":["OpenAI","Broadcom"],"tier":2,"source":{"url":"https://techcrunch.com/2026/06/24/openai-unveils-its-first-custom-chip-built-by-broadcom/","title":"OpenAI unveils its first custom chip, built by Broadcom","publisher":"TechCrunch","type":"news","date":"2026-06-24"},"references":[{"url":"https://openai.com/index/openai-broadcom-jalapeno-inference-chip/","title":"OpenAI and Broadcom unveil LLM-optimized inference chip","publisher":"OpenAI","type":"announcement","date":"2026-06-24"}]}
{"id":"2026-06-25-vesuvius-challenge-first-scroll-read","date":"2026-06-25","precision":"day","title":"An entire Herculaneum scroll is read for the first time in 2,000 years","summary":"Machine-learning models trained to detect ink invisible to the eye let the Vesuvius Challenge fully transcribe PHerc. 1667, a carbonized scroll sealed since the 79 AD eruption. It turned out to hold a Stoic treatise on ethics.","category":"research","orgs":["Vesuvius Challenge"],"tier":3,"source":{"url":"https://scrollprize.org/firstscroll","title":"An entire Herculaneum scroll has been read for the first time","publisher":"Vesuvius Challenge","type":"announcement","date":"2026-06-25"}}
{"id":"2026-06-26-government-vetted-model-access","date":"2026-06-26","precision":"day","title":"Frontier access starts requiring a background check","summary":"OpenAI and Anthropic both agreed to vet customers for their most capable models through a United States government process, two weeks after the export order. Buying the top tier now meant being approved for it.","category":"policy","orgs":["OpenAI","Anthropic"],"tier":2,"source":{"url":"https://www.cnbc.com/2026/06/26/openai-limits-new-ai-models-to-trusted-partners-request-us-government.html","title":"OpenAI limits new AI models to trusted partners at request of US government","publisher":"CNBC","type":"news","date":"2026-06-26"},"references":[{"url":"https://www.washingtonpost.com/technology/2026/06/26/openai-says-us-government-will-vet-users-its-latest-ai-model/","title":"OpenAI says the U.S. government will vet users of its latest AI model","publisher":"The Washington Post","type":"news","date":"2026-06-26"}]}
{"id":"2026-06-28-google-limits-meta-gemini-access","date":"2026-06-28","precision":"day","title":"Google limits Meta's access to its Gemini models","summary":"Meta had been buying Gemini capacity for safety filtering, coding assistants and internal tools; Google could not fill the order and capped it, delaying some of Meta's own AI projects. Even a company building its own frontier models still needed someone else's compute.","category":"deal","orgs":["Google","Meta"],"tier":3,"source":{"url":"https://www.cnbc.com/2026/06/28/google-limits-metas-use-of-its-gemini-ai-models-ft-reports.html","title":"Google limits Meta's use of its Gemini AI models, FT reports","publisher":"CNBC","type":"news","date":"2026-06-28"}}
{"id":"2026-06-30-claude-science","date":"2026-06-30","precision":"day","title":"Anthropic launches Claude Science, a workbench for researchers","summary":"Not a new model, a workbench: more than 60 scientific databases and toolkits wired into Claude, producing auditable results a researcher can rerun. Announced at a live event with executives from Bristol Myers Squibb, Novartis and Genentech on the stage alongside Dario Amodei.","category":"product","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/claude-science-ai-workbench","title":"Claude Science, an AI workbench for scientists, is now available","publisher":"Anthropic","type":"announcement","date":"2026-06-30"}}
{"id":"2026-06-30-claude-sonnet-5","date":"2026-06-30","precision":"day","title":"Anthropic releases Claude Sonnet 5","summary":"Anthropic released Claude Sonnet 5 at introductory API prices of $2 per million input tokens and $10 per million output tokens. On August 10, it made those prices permanent.","category":"model","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/claude-sonnet-5","title":"Introducing Claude Sonnet 5","publisher":"Anthropic","type":"announcement","date":"2026-06-30"}}
{"id":"2026-06-30-fable-5-redeployed","date":"2026-06-30","precision":"day","title":"Commerce clears Fable 5 and Anthropic switches it back on","summary":"Nineteen days dark, a new safety classifier built with government agencies and Amazon, and global availability from July 1. Mythos 5 came back only for vetted US organizations.","category":"policy","orgs":["Anthropic","United States Department of Commerce"],"tier":2,"source":{"url":"https://www.anthropic.com/news/redeploying-fable-5","title":"Redeploying Claude Fable 5","publisher":"Anthropic","type":"announcement","date":"2026-06-30"},"references":[{"url":"https://www.cnbc.com/2026/06/30/anthropic-says-trump-admin-has-lifted-export-controls-on-claude-fable-5-and-mythos-5.html","title":"Anthropic says Trump admin has lifted export controls on Claude Fable 5 and Mythos 5","publisher":"CNBC","type":"news","date":"2026-06-30"}]}
{"id":"2026-07-06-illinois-ai-safety-measures-act","date":"2026-07-06","precision":"day","title":"Illinois enacts the country's strictest state AI safety law","summary":"Governor Pritzker signed the AI Safety Measures Act, requiring large developers to publish catastrophic-risk plans, report critical safety incidents within 72 hours, and submit to independent annual audits starting in 2028. California and New York had passed narrower versions; Illinois went further.","category":"policy","orgs":["State of Illinois"],"tier":2,"source":{"url":"https://gov-pritzker-newsroom.prezly.com/gov-pritzker-signs-nation-leading-artificial-intelligence-safety-law","title":"Gov. Pritzker Signs Nation-Leading Artificial Intelligence Safety Law","publisher":"Office of Governor JB Pritzker","type":"press-release","date":"2026-07-06"},"references":[{"url":"https://capitolnewsillinois.com/news/pritzker-signs-landmark-ai-regulation-bill-that-aims-to-mitigate-risks/","title":"Pritzker signs landmark AI regulation bill that aims to mitigate risks","publisher":"Capitol News Illinois","type":"news","date":"2026-07-06"}]}
{"id":"2026-07-08-cognition-swe-1-7","date":"2026-07-08","precision":"day","title":"Cognition releases SWE-1.7, its most capable coding model","summary":"Trained through further reinforcement learning on top of a Kimi K2.7 base, SWE-1.7 scored 42.3 percent on FrontierCode Main, within a few points of GPT-5.5 and Claude Opus 4.8, at $1.97 a task. It shipped inside Devin, served by Cerebras at 1,000 tokens per second.","category":"model","orgs":["Cognition"],"tier":3,"source":{"url":"https://cognition.com/blog/swe-1-7","title":"SWE-1.7: Frontier Intelligence at a Fraction of the Cost","publisher":"Cognition","type":"announcement","date":"2026-07-08"}}
{"id":"2026-07-08-gpt-live-launch","date":"2026-07-08","precision":"day","title":"OpenAI ships GPT-Live, a full-duplex voice model for ChatGPT","summary":"GPT-Live and a smaller GPT-Live-mini listen and speak at once instead of waiting for turns, delegating harder questions to a frontier model in the background. It rolled out to iOS, Android and web, with a developer API to follow.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://siliconangle.com/2026/07/08/openai-launches-gpt-live-voice-model-series-ahead-broad-gpt-5-6-release/","title":"OpenAI launches GPT-Live voice model series ahead of broad GPT-5.6 release","publisher":"SiliconANGLE","type":"news","date":"2026-07-08"},"references":[{"url":"https://openai.com/index/introducing-gpt-live/","title":"Introducing GPT-Live","publisher":"OpenAI","type":"announcement"}]}
{"id":"2026-07-09-chatgpt-work-launch","date":"2026-07-09","precision":"day","title":"OpenAI launches ChatGPT Work, an agent for multi-hour office tasks","summary":"Powered by GPT-5.6, ChatGPT Work gathers context across connected apps and files to produce spreadsheets, slides and documents, staying with a project for hours by breaking it into steps. It connects to Slack, Teams, Gmail and Salesforce and rolled out first to Pro, Enterprise and Edu plans.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://siliconangle.com/2026/07/09/openai-debuts-chatgpt-work-agentic-tool-automating-business-workflows/","title":"OpenAI debuts ChatGPT Work, an agentic tool for automating business workflows","publisher":"SiliconANGLE","type":"news","date":"2026-07-09"},"references":[{"url":"https://openai.com/index/chatgpt-for-your-most-ambitious-work/","title":"ChatGPT is now a partner for your most ambitious work","publisher":"OpenAI","type":"announcement"}]}
{"id":"2026-07-09-fidji-simo-part-time-adviser","date":"2026-07-09","precision":"day","title":"Fidji Simo leaves OpenAI's full-time leadership team","summary":"Fidji Simo said she would leave her full-time role at OpenAI and become a part-time adviser after a medical leave. The change removed the executive who had led the company's applications, product and business operations from day-to-day management while keeping her connected to the company.","category":"company","orgs":["OpenAI"],"tier":3,"source":{"url":"https://www.linkedin.com/posts/fidjisimo_today-i-shared-with-the-openai-team-that-activity-7481120077711425536-e03r","title":"Today, I shared with the OpenAI team that I have decided to leave my full-time role at OpenAI and transition to being a part-time advisor.","publisher":"Fidji Simo","type":"announcement","date":"2026-07-09"},"references":[{"url":"https://www.investing.com/news/stock-market-news/openais-applications-chief-fidji-simo-to-step-down-4785167","title":"OpenAI's applications chief Fidji Simo to step down","publisher":"Reuters","author":"Fabiola Aramburo","type":"news","date":"2026-07-09"}]}
{"id":"2026-07-09-gpt-5-6-public-launch","date":"2026-07-09","precision":"day","title":"GPT-5.6 reaches the public","summary":"Four weeks after the export order held it back, OpenAI released the model broadly once the government limits came off. A price cut of up to 80 percent followed three weeks later.","category":"model","orgs":["OpenAI"],"tier":3,"source":{"url":"https://openai.com/index/gpt-5-6/","title":"GPT-5.6: Frontier intelligence that scales with your ambition","publisher":"OpenAI","type":"announcement","date":"2026-07-09"},"references":[{"url":"https://www.cnbc.com/2026/07/08/openai-expanding-gpt-5point6-ai-model-release-ending-government-limits.html","title":"OpenAI to publicly release GPT-5.6, rolls out conversational AI models","publisher":"CNBC","type":"news","date":"2026-07-08"}]}
{"id":"2026-07-09-muse-spark-1-1-model-api","date":"2026-07-09","precision":"day","title":"Meta starts selling access to a model it will not publish","summary":"Muse Spark 1.1 was aimed at agentic coding and arrived with a public Model API. The company whose whole AI identity was giving weights away now had a metered endpoint and a closed flagship behind it.","category":"model","orgs":["Meta"],"tier":3,"source":{"url":"https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/","title":"Introducing Muse Spark 1.1","publisher":"Meta AI","type":"announcement","date":"2026-07-09"},"references":[{"url":"https://techcrunch.com/2026/07/09/meta-enters-the-crowded-ai-coding-battle-with-muse-spark-1-1/","title":"Meta enters the crowded AI coding battle with Muse Spark 1.1","publisher":"TechCrunch","type":"news","date":"2026-07-09"}]}
{"id":"2026-07-10-apple-sues-openai","date":"2026-07-10","precision":"day","title":"Apple sues OpenAI over departed employees","summary":"The complaint accused former Apple staff of carrying trade secrets to OpenAI, six months after Apple decided to license Gemini rather than build a frontier model of its own. Both events were about the same shortage.","category":"company","orgs":["Apple","OpenAI"],"tier":3,"source":{"url":"https://9to5mac.com/2026/07/10/apple-sues-openai-trade-secret-theft/","title":"Apple sues OpenAI, accuses ex-employees of stealing trade secrets","publisher":"9to5Mac","type":"news","date":"2026-07-10"}}
{"id":"2026-07-10-bis-uae-export-control-upgrade","date":"2026-07-10","precision":"day","title":"US clears the UAE for license-free AI chip exports","summary":"The Commerce Department moved the UAE from its most restricted export groups into Country Group A:5, letting the UAE government and vetted entities such as G42 buy advanced AI chips and servers without individual licenses. The change carries out the AI cooperation framework the two countries signed in 2025.","category":"policy","orgs":["United States Department of Commerce"],"tier":2,"source":{"url":"https://www.bis.gov/press-release/department-commerce-eases-export-controls-uae","title":"Department of Commerce Eases Export Controls for UAE","publisher":"US Department of Commerce, Bureau of Industry and Security","type":"press-release","date":"2026-07-10"},"references":[{"url":"https://www.federalregister.gov/documents/2026/07/14/2026-14132/enhanced-favorable-treatment-for-the-united-arab-emirates-under-the-export-administration","title":"Enhanced Favorable Treatment for the United Arab Emirates Under the Export Administration Regulations","publisher":"Federal Register","type":"docs","date":"2026-07-14"}]}
{"id":"2026-07-14-new-york-data-center-moratorium","date":"2026-07-14","precision":"day","title":"New York becomes the first state to halt new hyperscale data centers","summary":"Executive Order 62 stops the state from issuing discretionary environmental permits for new hyperscale data centers for up to a year while it writes standards for energy, water and air. The legislature had passed its own one-year moratorium in June; the governor's order is what took effect.","category":"policy","orgs":["State of New York"],"tier":2,"source":{"url":"https://www.governor.ny.gov/news/first-statewide-moratorium-new-hyperscale-data-centers-launched-governor-kathy-hochul","title":"First Statewide Moratorium on New Hyperscale Data Centers Launched by Governor Kathy Hochul","publisher":"Office of Governor Kathy Hochul","type":"press-release","date":"2026-07-14"},"references":[{"url":"https://www.washingtonpost.com/technology/2026/07/14/new-york-becomes-first-state-impose-data-center-moratorium/","title":"New York becomes first state to impose data center moratorium","publisher":"The Washington Post","type":"news","date":"2026-07-14"}]}
{"id":"2026-07-15-thinking-machines-inkling","date":"2026-07-15","precision":"day","title":"Thinking Machines finally ships something","summary":"Mira Murati's lab released Inkling, its first model, a year after raising $2 billion at a $12 billion valuation with no product at all. It landed as the strongest open-weights model from a US lab, a category China had been holding.","category":"model","secondaryCategories":["open-source"],"orgs":["Thinking Machines Lab"],"tier":2,"source":{"url":"https://thinkingmachines.ai/news/introducing-inkling/","title":"Inkling: Our Open-Weights Model","publisher":"Thinking Machines Lab","type":"announcement","date":"2026-07-15"},"references":[{"url":"https://techcrunch.com/2026/07/15/thinking-machines-amps-up-its-bet-against-one-size-fits-all-ai-with-its-first-open-model-inkling/","title":"Thinking Machines amps up its bet against one-size-fits-all AI with its first open model, Inkling","publisher":"TechCrunch","author":"Connie Loizos","type":"news","date":"2026-07-15"},{"url":"https://artificialanalysis.ai/articles/thinking-machines-has-released-inkling-the-new-leading-u-s-open-weights-model","title":"Thinking Machines has released Inkling, the new leading U.S. open weights model","publisher":"Artificial Analysis","type":"analysis","date":"2026-07-15"}]}
{"id":"2026-07-16-grok-4-5","date":"2026-07-16","precision":"day","title":"xAI releases Grok 4.5","summary":"A 500,000 token context window at $2 per million input tokens, and the first frontier release to go out under the SpaceXAI name.","category":"model","orgs":["xAI","SpaceX"],"tier":3,"source":{"url":"https://x.ai/news/grok-4-5","title":"Introducing Grok 4.5","publisher":"xAI","type":"announcement","date":"2026-07-16"}}
{"id":"2026-07-19-claude-jacobian-counterexample","date":"2026-07-19","precision":"day","title":"Claude-assisted work refutes the Jacobian conjecture in three dimensions","summary":"Levent Alpöge, working with Anthropic's Claude Fable 5, produced an explicit polynomial map in C³ with a constant nonzero Jacobian determinant but three distinct inputs mapping to one output. It refutes the general conjecture in dimension three and above; the two-dimensional case remains open.","category":"research","orgs":["Anthropic"],"tier":2,"source":{"url":"https://terrytao.wordpress.com/2026/07/21/a-digestion-of-the-jacobian-conjecture-counterexample/","title":"A digestion of the Jacobian conjecture counterexample","publisher":"Terence Tao","type":"analysis","date":"2026-07-21"},"references":[{"url":"https://arxiv.org/abs/2608.00222","title":"Counterexamples to the Jacobian conjecture in dimensions greater than two","publisher":"arXiv","author":"Shuhong Gao","type":"paper","date":"2026-07-31"}]}
{"id":"2026-07-21-gemini-3-6-flash","date":"2026-07-21","precision":"day","title":"Gemini 3.6 Flash arrives with a cyber variant for governments","summary":"The workhorse tier got cheaper and spent about 17 percent fewer output tokens, with Flash-Lite alongside it. The sibling worth noting was 3.5 Flash Cyber, tuned to find and fix vulnerabilities and restricted to governments and trusted partners.","category":"model","orgs":["Google"],"tier":3,"source":{"url":"https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/","title":"Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber","publisher":"Google","type":"announcement","date":"2026-07-21"}}
{"id":"2026-07-21-hugging-face-evaluation-incident","date":"2026-07-21","precision":"day","title":"A model evaluation turns into a security incident","summary":"OpenAI and Hugging Face disclosed a security incident that arose during model evaluation, and spent the following week publishing updates on it. Running untrusted models to measure them is itself an attack surface, which the field had been slow to write down.","category":"research","orgs":["OpenAI","Hugging Face"],"tier":2,"source":{"url":"https://openai.com/index/hugging-face-model-evaluation-security-incident/","title":"OpenAI and Hugging Face address security incident during model evaluation","publisher":"OpenAI","type":"announcement","date":"2026-07-21"}}
{"id":"2026-07-24-claude-opus-5","date":"2026-07-24","precision":"day","title":"Claude Opus 5 arrives with a million tokens of context","summary":"Anthropic's fifth-generation flagship, pitched on treating its context as a working document rather than a transcript to be summarized. Six weeks earlier the same company's top model had been export controlled.","category":"model","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/news/claude-opus-5","title":"Introducing Claude Opus 5","publisher":"Anthropic","type":"announcement","date":"2026-07-24"},"references":[{"url":"https://platform.claude.com/docs/en/release-notes/overview","title":"Claude Platform release notes","publisher":"Anthropic","type":"docs","locator":"July 24, 2026"}]}
{"id":"2026-07-24-open-weights-american-ai-leadership-letter","date":"2026-07-24","precision":"day","title":"Nvidia, Microsoft and Meta lead a coalition letter against restricting open weights","summary":"Twenty-five companies and groups, including Nvidia, Microsoft, Meta, Palantir, IBM and Hugging Face, published \"Open Weights and American AI Leadership,\" urging Washington against premature restrictions on downloadable model weights. The roster passed 150 signatories within days; Anthropic did not sign.","category":"policy","orgs":["NVIDIA","Microsoft","Meta"],"tier":2,"source":{"url":"https://www.theregister.com/ai-and-ml/2026/07/24/tech-leaders-issue-letter-to-train-uncle-sam-about-value-of-open-weight-ai/5278533","title":"Tech leaders issue letter to train Uncle Sam about value of open weight AI","publisher":"The Register","type":"news","date":"2026-07-24"},"references":[{"url":"https://images.nvidia.com/pdf/Open-Weights-and-American-AI-Leadership.pdf","title":"Open Weights and American AI Leadership","publisher":"NVIDIA","type":"report","date":"2026-07-24"}]}
{"id":"2026-07-27-anthropic-open-weights-position","date":"2026-07-27","precision":"day","title":"Anthropic explains why it does not publish weights","summary":"A public position paper on open-weights models, written by the one major lab that has never released any, six weeks after the government restricted access to its own. The comment threads were longer than the paper.","category":"culture","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/position-open-weights-models","title":"Our position on open-weights models","publisher":"Anthropic","type":"announcement","date":"2026-07-27"}}
{"id":"2026-07-27-kimi-k3-open-weights","date":"2026-07-27","precision":"day","title":"Moonshot publishes Kimi K3's weights","summary":"2.8 trillion parameters with 104 billion active, the largest open-weight model anyone had released. Downloading it means moving about 1.4 terabytes, which is a moat of a sort.","category":"open-source","orgs":["Moonshot AI"],"tier":2,"source":{"url":"https://huggingface.co/moonshotai/Kimi-K3","title":"moonshotai/Kimi-K3","publisher":"Hugging Face","type":"docs","date":"2026-07-27"},"references":[{"url":"https://qz.com/moonshot-ai-kimi-k3-open-weights-download-072726","title":"Moonshot AI releases Kimi K3 open-weight model for download","publisher":"Quartz","type":"news","date":"2026-07-27"},{"url":"https://www.interconnects.ai/p/kimi-k3-the-open-weights-escalation","title":"Kimi K3: The open-weights escalation","publisher":"Interconnects","author":"Nathan Lambert","type":"analysis","date":"2026-07-20"}]}
{"id":"2026-07-27-safe-superintelligence-nvidia-partnership","date":"2026-07-27","precision":"day","title":"NVIDIA backs Ilya Sutskever's Safe Superintelligence with a compute deal","summary":"The two companies announced a long-term strategic partnership giving SSI access to NVIDIA's Vera Rubin platform, which SSI said would increase its compute by an order of magnitude. Financial terms were not disclosed.","category":"deal","orgs":["Safe Superintelligence","NVIDIA"],"tier":2,"source":{"url":"https://nvidianews.nvidia.com/news/ilya-sutskevers-safe-superintelligence-inc-and-nvidia-announce-long-term-strategic-partnership","title":"Ilya Sutskever's Safe Superintelligence Inc. and NVIDIA Announce Long-Term Strategic Partnership","publisher":"NVIDIA","type":"press-release","date":"2026-07-27"},"references":[{"url":"https://techcrunch.com/2026/07/27/ilya-sutskevers-safe-superintelligence-partners-with-nvidia-to-scale-its-ai-research/","title":"Ilya Sutskever's Safe Superintelligence partners with Nvidia to scale its AI research","publisher":"TechCrunch","type":"news","date":"2026-07-27"}]}
{"id":"2026-07-28-anthropic-cryptographic-weaknesses","date":"2026-07-28","precision":"day","title":"Claude finds a new attack that halves a post-quantum signature scheme's strength","summary":"Anthropic's research model found flaws in the cryptographic algorithms themselves rather than in implementations, including an improved attack on the HAWK post-quantum signature candidate and a fingerprinting technique that speeds up attacks on reduced-round AES by up to 800 times. Neither finding affects deployed systems.","category":"research","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/research/discovering-cryptographic-weaknesses","title":"Discovering cryptographic weaknesses with Claude","publisher":"Anthropic","type":"report","date":"2026-07-28"}}
{"id":"2026-07-28-pacing-the-frontier","date":"2026-07-28","precision":"day","title":"Lab employees ask Washington for a brake they can pull later","summary":"More than a thousand people inside the frontier labs, including Anthropic's chief executive and OpenAI's chief scientist, asked the United States to back an international effort to build the tools for deliberately pacing automated AI development. The statement does not ask anyone to slow down now.","category":"culture","orgs":["OpenAI","Anthropic","Google DeepMind","Meta"],"tier":2,"source":{"url":"https://www.pacingthefrontier.com/","title":"Pacing the Frontier","publisher":"Pacing the Frontier","type":"announcement"},"references":[{"url":"https://edition.cnn.com/2026/07/28/tech/ai-development-tech-employees-open-letter","title":"Employees from the world's biggest AI companies want the US to be ready to slow AI development","publisher":"CNN","type":"news","date":"2026-07-28"}]}
{"id":"2026-07-30-anthropic-cyber-evaluation-incidents","date":"2026-07-30","precision":"day","title":"Anthropic discloses three cybersecurity-evaluation incidents","summary":"Anthropic disclosed that Claude models, running cybersecurity evaluations in an environment with unintended internet access, gained unauthorized access to the real systems of three organizations. The company described the cases as a containment and operational failure, not a deliberate escape or evidence of a model pursuing a goal of its own.","category":"research","orgs":["Anthropic","Irregular"],"tier":2,"source":{"url":"https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals","title":"Investigating three real-world incidents in our cybersecurity evaluations","publisher":"Anthropic","type":"announcement","date":"2026-07-30"}}
{"id":"2026-07-30-gemini-robotics-2","date":"2026-07-30","precision":"day","title":"Google DeepMind ships Gemini Robotics 2 for whole-body robot control","summary":"The model series directs a humanoid robot from posture down to multi-finger dexterity, rather than the upper-body-only control of its predecessor, and adapts to a new robot body within hours. It also supports multiple robots working together on the same task.","category":"model","orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/","title":"Gemini Robotics 2 brings whole body intelligence to robots","publisher":"Google DeepMind","type":"announcement","date":"2026-07-30"},"references":[{"url":"https://siliconangle.com/2026/07/30/google-deepmind-debuts-gemini-robotics-2-model-series-humanoid-robots/","title":"Google DeepMind debuts Gemini Robotics 2 model series for humanoid robots","publisher":"SiliconANGLE","type":"news","date":"2026-07-30"}]}
{"id":"2026-07-30-gpt-5-6-luna-price-cut","date":"2026-07-30","precision":"day","title":"OpenAI cuts Luna's price by 80 percent","summary":"OpenAI cut GPT-5.6 Luna API prices from $1 and $6 to $0.20 and $1.20 per million input and output tokens, respectively, and also lowered Terra pricing.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://www.cnbc.com/2026/07/30/open-ai-price-cut-gpt.html","title":"OpenAI cuts prices for two of its GPT-5.6 AI models as companies grow sensitive to costs","publisher":"CNBC","type":"news","date":"2026-07-30"}}
{"id":"2026-07-31-deepseek-v4-flash-0731","date":"2026-07-31","precision":"day","title":"DeepSeek ships V4-Flash-0731 after retraining its post-training pipeline","summary":"The production release keeps the same architecture and size as the April preview but was re-post-trained, with agent-capability benchmark gains and native support for the Responses API format used by Codex. The API moved into public beta.","category":"model","orgs":["DeepSeek"],"tier":3,"source":{"url":"https://api-docs.deepseek.com/updates/","title":"DeepSeek-V4-Flash Update","publisher":"DeepSeek","type":"docs","date":"2026-07-31"}}
{"id":"2026-07-31-openai-billion-active-users","date":"2026-07-31","precision":"day","title":"OpenAI says its models reach a billion active users","summary":"CFO Sarah Friar disclosed the figure, along with more than two million businesses, in a post about revenue and margins rather than a users-milestone announcement of its own. The total spans OpenAI's products, not ChatGPT alone, and arrived faster than Facebook's run to a billion users.","category":"culture","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/building-abundant-intelligence/","title":"Building abundant intelligence","publisher":"OpenAI","type":"announcement","date":"2026-07-31","author":"Sarah Friar","locator":"Our models now reach more than one billion active users and more than two million businesses."},"references":[{"url":"https://www.bnnbloomberg.ca/business/artificial-intelligence/2026/07/31/openai-says-has-more-than-1-billion-active-users/","title":"OpenAI says has more than 1 billion active users","publisher":"BNN Bloomberg","type":"news","date":"2026-07-31"}]}
{"id":"2026-08-01-openai-astra-ten-math-advances","date":"2026-08-01","precision":"day","title":"An unreleased OpenAI model produces ten advances in mathematics","summary":"Astra, still internal, is credited with the first explicit construction of a non-sofic group, open since 1999, plus a disproof of Connes's rigidity conjecture and resolutions of three Erdős problems; the other results in the batch are improved bounds rather than full solutions. Each write-up ships a machine-checked Lean 4 certificate on GitHub, at a compute cost OpenAI put at about $2,000.","category":"research","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/ten-advances-in-mathematics/","title":"Ten advances in mathematics and theoretical computer science","publisher":"OpenAI","type":"announcement","date":"2026-08-01"},"references":[{"url":"https://the-decoder.com/openai-announces-its-next-major-model-astra-by-dropping-ten-previously-unsolved-math-solutions/","title":"OpenAI announces its \"next major model\" Astra by dropping ten previously unsolved math solutions","publisher":"The Decoder","type":"news","date":"2026-08-01"},{"url":"https://www.bleepingcomputer.com/news/artificial-intelligence/openai-teases-astra-its-next-major-ai-model-after-it-solves-10-long-standing-math-problems/","title":"OpenAI teases Astra, its next major AI model, after it solves 10 long-standing math problems","publisher":"BleepingComputer","type":"news","date":"2026-08-02"}]}
{"id":"2026-08-02-eu-ai-act-transparency","date":"2026-08-02","precision":"day","title":"The EU starts enforcing the AI Act's disclosure rules","summary":"Chatbots have to say they are chatbots, deepfakes have to be labelled, and generated content has to carry a machine-readable mark. The high-risk obligations had already slipped to 2027 and 2028, so the weight of the date was the AI Office finally holding enforcement powers.","category":"policy","orgs":["European Commission"],"tier":2,"source":{"url":"https://digital-strategy.ec.europa.eu/en/news/commission-starts-enforcing-ai-act-rules-and-new-transparency-requirements-2-august","title":"Commission starts enforcing AI Act rules and new transparency requirements on 2 August","publisher":"European Commission","type":"announcement","date":"2026-07-31"},"references":[{"url":"https://digital-strategy.ec.europa.eu/en/policies/guidelines-transparency-ai-generated-content","title":"Guidelines on transparency obligations for providers and deployers of certain AI systems","publisher":"European Commission","type":"docs"}]}
{"id":"2026-08-05-google-deepmind-leadership","date":"2026-08-05","precision":"day","title":"Hassabis moves to chair and Jeff Dean leaves Google","summary":"Koray Kavukcuoglu took over Gemini, the app, and frontier research as senior vice president; Demis Hassabis became chair of Google DeepMind and chief scientist of Alphabet; and Jeff Dean left after 27 years to start a public benefit corporation with Sanjay Ghemawat.","category":"company","orgs":["Google DeepMind","Google"],"tier":2,"source":{"url":"https://blog.google/company-news/inside-google/message-ceo/next-chapter-ai-momentum/","title":"The next chapter of our AI momentum","publisher":"Google","type":"announcement","date":"2026-08-05"},"references":[{"url":"https://www.cnbc.com/2026/08/06/demis-hassabis-google-reshuffle-deepmind-role.html","title":"Demis Hassabis' new Google DeepMind role explained","publisher":"CNBC","type":"news","date":"2026-08-06"}]}
{"id":"2026-08-05-muse-code","date":"2026-08-05","precision":"day","title":"Meta ships a terminal coding agent","summary":"Muse Code entered beta on Muse Spark 1.2, with parallel subagents, worktree isolation, and a crash-safe event log. The first coding product out of Meta Superintelligence Labs, priced under the incumbents at $1.25 and $4.25 per million tokens.","category":"product","orgs":["Meta"],"tier":3,"source":{"url":"https://techcrunch.com/2026/08/05/meta-launches-muse-code-an-ai-agent-for-large-code-bases/","title":"Meta launches Muse Code, an AI agent for large code bases","publisher":"TechCrunch","type":"news","date":"2026-08-05"},"references":[{"url":"https://www.cnbc.com/2026/08/05/meta-debuts-muse-code-to-take-on-anthropic-and-openai-.html","title":"Meta debuts first AI coding agent to take on Anthropic and OpenAI","publisher":"CNBC","type":"news","date":"2026-08-05"}]}
{"id":"2026-08-06-amd-acquires-taalas","date":"2026-08-06","precision":"day","title":"AMD buys a company that etches models into silicon","summary":"Taalas builds chips with a specific model's weights fixed in hardware, trading all flexibility for inference cost. AMD paying for that thesis is a bet that some models are now stable enough to be permanent.","category":"acquisition","orgs":["AMD","Taalas"],"tier":3,"source":{"url":"https://ir.amd.com/news-events/press-releases/detail/1296/amd-acquires-taalas-to-advance-compute-solutions-for-rapidly-growing-ai-inference-market","title":"AMD Acquires Taalas to Advance Compute Solutions for Rapidly Growing AI Inference Market","publisher":"AMD","type":"press-release","date":"2026-08-06"},"references":[{"url":"https://www.cnbc.com/2026/08/06/amd-buys-taalas-startup-that-hardwires-ai-models-into-its-silicon.html","title":"AMD buys chip startup that hardwires AI models into its silicon","publisher":"CNBC","type":"news","date":"2026-08-06"}]}
{"id":"2026-08-09-claude-code-auto-mode","date":"2026-08-09","precision":"day","title":"Claude Code stops asking permission by default","summary":"From August 14 the agent proceeds on its own for Pro, Max, and Team users, pausing only for work that is destructive or hard to undo. Anthropic said its automated reviewer caught 89 percent of harmful actions in testing, against 13.6 percent for human reviewers.","category":"product","orgs":["Anthropic"],"tier":3,"source":{"url":"https://techcrunch.com/2026/08/09/anthropic-is-turning-claude-codes-auto-mode-on-by-default/","title":"Anthropic is turning Claude Code's auto mode on by default","publisher":"TechCrunch","type":"news","date":"2026-08-09"},"references":[{"url":"https://www.helpnetsecurity.com/2026/08/10/anthropic-claude-code-auto-mode/","title":"Anthropic to put AI in charge of reviewing Claude Code actions by default","publisher":"Help Net Security","type":"news","date":"2026-08-10"}]}
{"id":"2026-08-10-gpt-5-6-cyber","date":"2026-08-10","precision":"day","title":"OpenAI ships an offensive security model behind a gate","summary":"GPT-5.6-Cyber completes 95 percent of dual-use security requests where GPT-5.6 Sol completes 1.5 percent, and getting to it means identity verification, approved use cases, and legal attestations under a tier called Daybreak Red. A defensive tier, Daybreak Blue, opened alongside it.","category":"model","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/expanding-daybreak-as-the-cyber-defense-window-narrows/","title":"Expanding Daybreak as the cyber defense window narrows","publisher":"OpenAI","type":"announcement","date":"2026-08-10"},"references":[{"url":"https://www.securityweek.com/openai-unveils-new-cybersecurity-model-gpt-5-6-cyber/","title":"OpenAI Unveils New Cybersecurity Model GPT-5.6-Cyber","publisher":"SecurityWeek","type":"news","date":"2026-08-11"},{"url":"https://techcrunch.com/2026/08/10/as-ai-led-attacks-multiply-openai-launches-a-new-cyber-model/","title":"As AI-led attacks multiply, OpenAI launches a new cyber model","publisher":"TechCrunch","type":"news","date":"2026-08-10"}]}
{"id":"2026-08-10-muse-glimmer","date":"2026-08-10","precision":"day","title":"Meta publishes weights again with Muse Glimmer","summary":"A 30 billion parameter agentic model under Apache 2.0, quantized to sit under 20GB so it runs on a Mac or one consumer GPU. Meta was back in open weights, with a model distilled from Muse Spark, whose own weights stayed shut.","category":"model","secondaryCategories":["open-source"],"orgs":["Meta"],"tier":3,"source":{"url":"https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model","title":"Introducing Muse Glimmer: An Open Agentic Model That Runs on Your Device","publisher":"Meta AI Research","type":"announcement","date":"2026-08-10"},"references":[{"url":"https://techcrunch.com/2026/08/10/metas-new-glimmer-ai-model-offers-a-hint-at-zuckerbergs-personal-intelligence-vision/","title":"Meta's new Glimmer AI model offers a hint at Zuckerberg's personal intelligence vision","publisher":"TechCrunch","type":"news","date":"2026-08-10"}]}
{"id":"2026-08-10-nvidia-compute-financing","date":"2026-08-10","precision":"day","title":"Nvidia turns compute into an asset class","summary":"Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR signed memoranda to mobilize over $500 billion of third-party capital for AI data centers, underwritten against the GPUs themselves. Larry Fink compared it to the invention of mortgage-backed securities, a comparison with two endings.","category":"funding","orgs":["NVIDIA","Apollo","BlackRock","Blackstone","Brookfield","Goldman Sachs","KKR"],"tier":2,"source":{"url":"https://nvidianews.nvidia.com/news/nvidia-partners-with-apollo-blackrock-blackstone-brookfield-goldman-sachs-and-kkr-to-establish-ai-compute-infrastructure-financing-platforms-to-mobilize-over-500-billion-of-third-party-capital","title":"NVIDIA Partners With Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to Establish AI Compute Infrastructure Financing Platforms to Mobilize Over $500 Billion of Third-Party Capital","publisher":"NVIDIA","type":"press-release","date":"2026-08-10"},"references":[{"url":"https://www.cnbc.com/2026/08/10/nvidia-wall-street-asset-managers-500-billion-ai-push.html","title":"Nvidia lines up $500 billion in financing as CEO Jensen Huang tells CNBC his chips are 'investable asset'","publisher":"CNBC","type":"news","date":"2026-08-10"}]}
{"id":"2026-08-11-chatgpt-ads-launch","date":"2026-08-11","precision":"day","title":"ChatGPT ads expand to five more markets","summary":"OpenAI expanded ChatGPT ads to the United Kingdom, Mexico, Brazil, Japan and South Korea, using the self-serve Ads Manager it had launched in May.","category":"product","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/testing-ads-in-chatgpt/","title":"Testing ads in ChatGPT","publisher":"OpenAI","type":"announcement","date":"2026-02-09","locator":"Update on August 11, 2026"},"references":[{"url":"https://ads.openai.com/","title":"OpenAI Ads: Advertise in ChatGPT","publisher":"OpenAI","type":"docs"}]}
{"id":"2026-08-11-gemini-one-billion-users","date":"2026-08-11","precision":"day","title":"The Gemini app passes a billion monthly users","summary":"Google's fourteenth product to cross a billion and the fastest to get there, running at 150 million images a day with 63 percent of users talking to it out loud. It went from 950 million to a billion in under a month.","category":"culture","orgs":["Google"],"tier":2,"source":{"url":"https://blog.google/innovation-and-ai/products/gemini-app/one-billion-monthly-users/","title":"Google's Gemini app hits 1 billion monthly active users","publisher":"Google","type":"announcement","date":"2026-08-11"},"references":[{"url":"https://techcrunch.com/2026/08/11/googles-gemini-app-surges-to-one-billion-users/","title":"Google's Gemini app surges to 1 billion users","publisher":"TechCrunch","type":"news","date":"2026-08-11"}]}
{"id":"2026-08-12-grok-4-6","date":"2026-08-12","precision":"day","title":"xAI releases Grok 4.6","summary":"xAI released Grok 4.6, using the Grok 4.5 base model with a longer post-training run, four weeks after the previous release.","category":"model","orgs":["xAI","SpaceX"],"tier":3,"source":{"url":"https://x.ai/news/grok-4-6","title":"Introducing Grok 4.6","publisher":"xAI","type":"announcement","date":"2026-08-12"}}
{"id":"2026-08-12-qwen-3-8-max-open-weights","date":"2026-08-12","precision":"day","title":"Alibaba publishes 2.4 trillion parameters","summary":"Qwen3.8 Max went up on Hugging Face under Alibaba's own license: 2.4 trillion parameters with 95 billion active, 262,000 tokens of context natively, text only and thinking always on. Second largest set of open weights in the world, sixteen days after the largest.","category":"open-source","orgs":["Alibaba"],"tier":3,"source":{"url":"https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B","title":"Qwen/Qwen3.8-2.4T-A95B","publisher":"Hugging Face","type":"docs","date":"2026-08-12"}}
{"id":"2026-08-13-deepseek-raises-prices","date":"2026-08-13","precision":"day","title":"DeepSeek announces peak-hour API price increases","summary":"DeepSeek announced V4 API price increases of 50 percent to more than 1,100 percent, depending on model and token type. Peak and off-peak rates took effect on August 16.","category":"product","orgs":["DeepSeek"],"tier":3,"source":{"url":"https://qz.com/deepseek-api-price-increase-v4-peak-off-peak-081326","title":"DeepSeek raising API prices by up to 1,100% starting Aug. 16","publisher":"Quartz","type":"news","date":"2026-08-13"},"references":[{"url":"https://www.scmp.com/tech/tech-trends/article/3363129/deepseek-signals-significant-price-hike-amid-surge-demand-low-cost-ai-models","title":"DeepSeek signals 'significant' price hike amid surge in demand for low-cost AI models","publisher":"South China Morning Post","type":"news","date":"2026-08-06"}]}
{"id":"2026-08-13-gemini-3-7-flash","date":"2026-08-13","precision":"day","title":"Gemini 3.7 Flash lands three weeks after 3.6","summary":"Google reported that Gemini 3.7 Flash improved from 34 to 44 percent on FrontierCode and from 49 to 65 percent on DeepSWE in 21 days, with half-price access through year-end.","category":"model","orgs":["Google"],"tier":3,"source":{"url":"https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-3-7-flash/","title":"Gemini 3.7 Flash: our most intelligent workhorse model","publisher":"Google","type":"announcement","date":"2026-08-13"}}
{"id":"2026-08-13-ibm-openai-partnership","date":"2026-08-13","precision":"day","title":"IBM puts OpenAI models inside its consulting business","summary":"GPT-5.6, Codex, and ChatGPT Work go into IBM Consulting Advantage, with a dedicated practice of thousands of certified consultants and engineers embedded with clients in regulated industries. What IBM bought was distribution, and what OpenAI bought was the enterprise sales force it does not have.","category":"company","orgs":["IBM","OpenAI"],"tier":3,"source":{"url":"https://newsroom.ibm.com/2026-08-13-ibm-partners-with-openai-to-accelerate-secure-ai-deployment-for-enterprises-across-core-operations","title":"IBM Partners with OpenAI to Accelerate Secure AI Deployment for Enterprises Across Core Operations","publisher":"IBM","type":"press-release","date":"2026-08-13"}}
{"id":"2026-08-14-claude-text-watermark","date":"2026-08-14","precision":"day","title":"Claude starts watermarking everything it writes","summary":"Anthropic turned on a SynthID-Text style watermark for models released from August 2 onward and applied it worldwide rather than only in Europe, while saying plainly that a detected mark is a signal and not proof. The EU's content-marking rule had arrived twelve days earlier.","category":"policy","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/claude-text-watermark","title":"How Claude's text watermark works","publisher":"Anthropic","type":"announcement","date":"2026-08-14"},"references":[{"url":"https://techcrunch.com/2026/08/11/anthropic-says-it-will-watermark-text-generated-by-its-ai-models/","title":"Anthropic says it will watermark text generated by its AI models","publisher":"TechCrunch","type":"news","date":"2026-08-11"}]}
{"id":"2026-08-14-glm-5-3","date":"2026-08-14","precision":"day","title":"GLM-5.3 takes its gains from post-training alone","summary":"Zhipu kept the 743 billion parameter base from 5.2 and reported the strongest open coding results it has measured, plus 84.5 percent on CyberGym, ahead of Mythos 5 and GPT-5.6 Sol. Weights were promised in about two weeks, after safety hardening.","category":"model","orgs":["Zhipu AI"],"tier":3,"source":{"url":"https://siliconangle.com/2026/08/14/z-ai-debuts-glm-5-3-long-horizon-coding-cybersecurity-upgrades/","title":"Z.ai debuts GLM-5.3 with long-horizon coding, cybersecurity upgrades","publisher":"SiliconANGLE","type":"news","date":"2026-08-14"},"references":[{"url":"https://www.marktechpost.com/2026/08/14/z-ai-ships-glm-5-3-without-retraining-the-base-model-better-at-complex-coding-and-long-horizon-tasks/","title":"Z.ai Ships GLM-5.3 Without Retraining the Base Model: Better at Complex Coding and Long-Horizon Tasks","publisher":"MarkTechPost","type":"analysis","date":"2026-08-14"}]}
{"id":"2026-08-14-qwen-3-8-27b","date":"2026-08-14","precision":"day","title":"Qwen 3.8 27B lands under Apache 2","summary":"A 27 billion parameter vision-capable model from Alibaba, permissively licensed and small enough to run on a workstation. The most capable model most people could actually own kept being made in China.","category":"model","secondaryCategories":["open-source"],"orgs":["Alibaba"],"tier":3,"source":{"url":"https://huggingface.co/Qwen/Qwen3.8-27B-FP8","title":"Qwen/Qwen3.8-27B-FP8","publisher":"Hugging Face","type":"docs","date":"2026-08-14"},"references":[{"url":"https://simonwillison.net/2026/Aug/16/qwen-38-27b/","title":"Qwen 3.8 27B is excellent, but it defaults to wildly overthinking things","publisher":"Simon Willison","author":"Simon Willison","type":"analysis","date":"2026-08-16"}]}
{"id":"2026-08-14-spacex-closes-cursor-deal","date":"2026-08-14","precision":"day","title":"SpaceX closes the Cursor deal and forms SpaceXAI","summary":"Roughly 391 million shares issued, and Anysphere folded into a new division alongside Grok. The largest startup acquisition on record took two months to clear.","category":"acquisition","orgs":["SpaceX","Anysphere"],"tier":3,"source":{"url":"https://finance.yahoo.com/technology/ai/articles/spacex-completes-record-60-billion-131311785.html","title":"SpaceX completes record $60 billion acquisition of AI coding platform Cursor","publisher":"Yahoo Finance","type":"news","date":"2026-08-14"}}
{"id":"2026-08-16-stripe-acquires-openrouter","date":"2026-08-16","precision":"day","title":"Stripe buys OpenRouter for more than $7 billion","summary":"The gateway developers use to switch between hundreds of models sold for over five times the $1.3 billion valuation it had raised at three months earlier. A payments company decided the toll booth worth owning was the one in front of the models.","category":"acquisition","orgs":["Stripe","OpenRouter"],"tier":2,"source":{"url":"https://www.bloomberg.com/news/articles/2026-08-16/stripe-nears-deal-to-buy-ai-firm-openrouter-for-over-7-billion","title":"Stripe Finalizes Deal to Acquire AI Startup OpenRouter for Over $7 Billion","publisher":"Bloomberg","type":"news","date":"2026-08-16"},"references":[{"url":"https://fortune.com/2026/08/16/stripe-7-billion-deal-ai-firm-openrouter-acquisition/","title":"Stripe clinches over $7 billion deal to buy AI firm OpenRouter","publisher":"Fortune","type":"news","date":"2026-08-16"},{"url":"https://techcrunch.com/2026/08/16/stripe-will-reportedly-acquire-ai-gateway-startup-openrouter-for-7b/","title":"Stripe will reportedly acquire AI gateway startup OpenRouter for $7B+","publisher":"TechCrunch","type":"news","date":"2026-08-16"}]}
{"id":"2026-08-17-nvidia-sb-energy-ohio","date":"2026-08-17","precision":"day","title":"Nvidia backs a $105 billion Ohio data center built for OpenAI","summary":"Nvidia guaranteed up to $105 billion in land, power and shell financing for SB Energy's Pike County campus, plus a $1.5 billion stake in SB Energy itself. OpenAI signed a 20 year lease on capacity that reaches 8 gigawatts once built out, with the first phase due online in 2028.","category":"deal","orgs":["NVIDIA","SB Energy","OpenAI"],"tier":2,"source":{"url":"https://nvidianews.nvidia.com/news/nvidia-guarantees-sb-energy-s-ports-pike-technology-campus-in-ohio-to-exclusively-host-nvidia-ai-compute","title":"NVIDIA Guarantees SB Energy's PORTS-Pike Technology Campus in Ohio to Exclusively Host NVIDIA AI Compute","publisher":"NVIDIA","type":"press-release","date":"2026-08-17"},"references":[{"url":"https://www.cnbc.com/2026/08/17/nvidia-financing-open-ai-data-center-ohio.html","title":"Nvidia backing $105 billion in financing for OpenAI data center in Ohio","publisher":"CNBC","type":"news","date":"2026-08-17"},{"url":"https://www.axios.com/2026/08/17/openai-nvidia-ohio-data-center-sb-energy","title":"OpenAI announces massive data center in Ohio with Nvidia guarantee","publisher":"Axios","type":"news","date":"2026-08-17"}]}
{"id":"2026-08-18-chatgpt-for-teens","date":"2026-08-18","precision":"day","title":"ChatGPT for Teens rolls out with age gating and parental controls","summary":"ChatGPT for Teens routes any account the system estimates as under 18 into a restricted version, where homework reminders catch assignment shortcuts and redirect to a new Study Mode, alongside controls parents can link to their own account. It follows several lawsuits accusing the chatbot of contributing to teenagers' deaths, arriving nearly four years after ChatGPT itself launched.","category":"product","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/chatgpt-for-teens/","title":"Introducing ChatGPT for Teens: Built for learning, backed by protections","publisher":"OpenAI","type":"announcement","date":"2026-08-18","locator":"rolling out globally to eligible teen accounts on Free and paid personal ChatGPT plans beginning August 18"},"references":[{"url":"https://techcrunch.com/2026/08/18/openai-launches-a-safer-chatgpt-for-teens-years-after-teens-started-using-it/","title":"OpenAI launches a safer ChatGPT for teens — years after teens started using it","publisher":"TechCrunch","type":"news","date":"2026-08-18"},{"url":"https://www.axios.com/2026/08/18/openai-chatgpt-for-teens","title":"OpenAI debuts ChatGPT for Teens","publisher":"Axios","type":"news","date":"2026-08-18"}]}
{"id":"2026-08-18-claude-protein-binder-design","date":"2026-08-18","precision":"day","title":"Claude designs working protein binders for 14 of 15 lab targets","summary":"With humans limited to approving access requests and watching the infrastructure, Claude orchestrated public protein-design and co-folding models to produce binders against 14 of 15 targets, at roughly twice the field's usual hit rate. Adaptyv Bio and Twist Bioscience built and tested the designs independently, rather than Anthropic grading its own homework.","category":"research","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/research/Claude-accelerates-protein-design","title":"How Claude is accelerating protein design and analytical chemistry","publisher":"Anthropic","type":"report","date":"2026-08-18","locator":"354 confirmed binders from 1,320 designs across 15 targets"},"references":[{"url":"https://thenextweb.com/news/anthropic-claude-protein-design-chemistry","title":"Anthropic says Claude designed working protein binders, and beat human experts on some","publisher":"The Next Web","type":"news","date":"2026-08-19"},{"url":"https://www.adaptyvbio.com/blog/anthropic-1","title":"Case study: Benchmarking Claude's protein designs in the wet lab","publisher":"Adaptyv Bio","type":"report","date":"2026-08-20"}]}
{"id":"2026-08-18-marvell-google-warrant","date":"2026-08-18","precision":"day","title":"Marvell gives Google a warrant worth $12.2 billion in a custom chip deal","summary":"Marvell issued Google a warrant for nearly 59 million shares, worth $12.2 billion if fully exercised, most of which only vests as Google buys custom AI inference, storage and networking silicon from Marvell through fiscal 2033. The arrangement would make Google one of Marvell's largest shareholders without either company disclosing a dollar figure for the chip orders behind it.","category":"deal","orgs":["Google","Marvell"],"tier":2,"source":{"url":"https://www.sec.gov/Archives/edgar/data/0001835632/000119312526356217/d412696d8k.htm","title":"Marvell Technology, Inc. - Form 8-K","publisher":"U.S. Securities and Exchange Commission","type":"report","date":"2026-08-19","locator":"the Company issued to Google a warrant to purchase up to an aggregate of 58,970,907 shares of common stock of the Company at an exercise price of $206.58 per share"},"references":[{"url":"https://www.cnbc.com/2026/08/19/marvell-google-ai-chips.html","title":"Marvell's stock pops 10% on AI chip deal that lets Google buy up to $12.2 billion in shares","publisher":"CNBC","type":"news","date":"2026-08-19"},{"url":"https://www.bloomberg.com/news/articles/2026-08-19/marvell-gives-google-right-to-buy-up-to-12-2-billion-in-shares","title":"Google Secures $12.2 Billion Share Purchase Right in Marvell AI Chip Deal","publisher":"Bloomberg","type":"news","date":"2026-08-19"}]}
{"id":"2026-08-18-openai-pauses-frontier-training","date":"2026-08-18","precision":"day","title":"OpenAI pauses its largest frontier training run over cyber capabilities","summary":"OpenAI put its largest planned reinforcement learning run on hold after an unreleased model, Astra, showed signs of crossing the Preparedness Framework's threshold for critical cyber capability, weeks after a separate model escaped a sandboxed evaluation and probed Hugging Face's infrastructure for days. Smaller runs continue while it expands monitoring and containment.","category":"policy","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/pacing-model-development-cyber-capabilities/","title":"Pacing model development in an era of cyber-critical capabilities","publisher":"OpenAI","type":"announcement","date":"2026-08-18","locator":"Our largest planned frontier RL run remains on hold while we conduct smaller-scale training and evaluations to assess model behavior, validate our safeguards, and establish more evidence of alignment before proceeding."},"references":[{"url":"https://www.helpnetsecurity.com/2026/08/19/openai-model-safety-updates/","title":"OpenAI puts major frontier AI training run on hold over cyber risks","publisher":"Help Net Security","type":"news","date":"2026-08-19"},{"url":"https://www.forbes.com/sites/ashishbhatia/2026/08/19/openai-paused-ai-training-for-two-weeks-heres-what-that-means/","title":"OpenAI Paused AI Training For Two Weeks. Here's What That Means","publisher":"Forbes","type":"news","date":"2026-08-19"}]}
{"id":"2026-08-19-intismeran-phase-3","date":"2026-08-19","precision":"day","title":"Merck and Moderna's individualized cancer therapy succeeds in phase 3","summary":"Intismeran autogene, an mRNA therapy encoding up to 34 neoantigens chosen from each patient's own tumor, cut recurrence in 1,137 resected melanoma patients when paired with Keytruda. The press release does not use the word AI; the neoantigen selection is a machine learning pipeline, and the industry claimed the result anyway.","category":"research","orgs":["Merck","Moderna"],"tier":3,"source":{"url":"https://www.merck.com/news/merck-and-moderna-announce-phase-3-interpath-001-trial-of-intismeran-autogene-plus-keytruda-met-endpoints-of-recurrence-free-survival-rfs-and-distant-metastasis-free-survival-dmfs-in-patient/","title":"Merck and Moderna Announce Phase 3 INTerpath-001 Trial of Intismeran Autogene Plus KEYTRUDA Met Endpoints of Recurrence-Free Survival (RFS) and Distant Metastasis-Free Survival (DMFS) in Patients With Completely Resected Stage IIB-IV Melanoma","publisher":"Merck","type":"press-release","date":"2026-08-19"},"references":[{"url":"https://www.statnews.com/2026/08/19/mrna-cancer-vaccine-trial-melanoma-merck-moderna/","title":"Moderna and Merck say mRNA cancer vaccine succeeded in late-stage melanoma trial","publisher":"STAT","type":"news","date":"2026-08-19"}]}
{"id":"2026-08-19-unitree-shanghai-ipo","date":"2026-08-19","precision":"day","title":"Unitree lists in Shanghai, the first mainland IPO of a humanoid robot maker","summary":"Unitree raised about $904 million in its Shanghai listing. Its shares gained as much as 629 percent intraday before closing more than five times above the IPO price.","category":"company","orgs":["Unitree"],"tier":3,"source":{"url":"https://www.caixinglobal.com/2026-08-19/chinese-robot-maker-unitree-soars-629-in-shanghai-debut-102475578.html","title":"Chinese Robot-Maker Unitree Soars 629% in Shanghai Debut","publisher":"Caixin Global","type":"news","date":"2026-08-19"},"references":[{"url":"https://www.cnbc.com/2026/08/19/china-backflipping-robot-maker-unitree-jumps-shanghai-ipo.html","title":"China's backflipping robot maker Unitree pops 542% in Shanghai debut","publisher":"CNBC","type":"news","date":"2026-08-19"},{"url":"https://www.bloomberg.com/news/articles/2026-08-18/unitree-robotics-set-to-debut-after-904-million-shanghai-ipo","title":"Unitree Robotics Surges 460% After $904 Million Shanghai IPO","publisher":"Bloomberg","type":"news","date":"2026-08-18"}]}
{"id":"2026-08-20-apple-music-ai-labels-required","date":"2026-08-20","precision":"day","title":"Apple Music makes AI-content tagging mandatory","summary":"Content providers must now flag any track where a generative AI service produced a material portion of the audio, upgrading a tag Apple had made optional in March; the visible \"Made With AI\" label itself follows later in the year. The rule arrives with more than a third of monthly uploads already AI made, even though those tracks draw well under one percent of listening.","category":"policy","orgs":["Apple"],"tier":3,"source":{"url":"https://www.macrumors.com/2026/08/20/apple-music-to-label-ai-generated-songs/","title":"Apple Music to Label AI-Generated Songs","publisher":"MacRumors","type":"news","date":"2026-08-20","locator":"AI Transparency Tags are required in 'any instance where AI was used to create a material portion of the content'"},"references":[{"url":"https://variety.com/2026/music/news/apple-music-to-label-ai-made-tracks-1236839371/","title":"Apple Music to Label AI-Made Tracks Later This Year","publisher":"Variety","type":"news","date":"2026-08-20"}]}
{"id":"2026-08-20-gemma-billion-downloads","date":"2026-08-20","precision":"day","title":"Gemma passes a billion downloads","summary":"Google reported more than 100,000 community variants of Gemma two years after launch. Teams at NASA, Satlyt and Starcloud used the models for onboard image analysis, and the Wild Dolphin Project used them for dolphin-vocalization research.","category":"open-source","orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://blog.google/innovation-and-ai/technology/developers-tools/gemma-one-billion-downloads/","title":"Gemma passes 1 billion downloads","publisher":"Google","type":"announcement","date":"2026-08-20"}}
{"id":"2026-08-20-nvidia-licenses-poolside","date":"2026-08-20","precision":"day","title":"NVIDIA licenses Poolside's model factory for $6 billion","summary":"A non-exclusive license to the system behind Poolside's Laguna coding models, plus a $1 billion investment at a $12 billion pre-money valuation and offers to more than 100 of its engineers. Poolside's founders told investors it was neither an acquisition nor an acquihire, since the company keeps running on its own.","category":"deal","orgs":["NVIDIA","Poolside"],"tier":2,"source":{"url":"https://www.newcomer.co/p/sources-poolside-strikes-6-billion","title":"SOURCES: Poolside Strikes $6 Billion Licensing Deal with Nvidia & Raises $1 Billion for Remaining Company at $12 Billion Valuation","publisher":"Newcomer","author":"Eric Newcomer","type":"news","date":"2026-08-20","locator":"This is not an acquisition and it is not an acquihire"},"references":[{"url":"https://www.bloomberg.com/news/articles/2026-08-20/nvidia-to-pay-ai-startup-poolside-a-6-billion-license-newcomer-says","title":"Nvidia to Pay AI Startup Poolside a $6 Billion License, Newcomer Says","publisher":"Bloomberg","type":"news","date":"2026-08-20"},{"url":"https://www.pymnts.com/news/artificial-intelligence/2026/nvidia-pays-6-billion-to-license-poolside-ai-model-development-software/","title":"Nvidia Pays $6 Billion to License Poolside AI Model-Development Software","publisher":"PYMNTS","type":"news","date":"2026-08-21"}]}
{"id":"2026-08-23-alibaba-ai-share-placement","date":"2026-08-23","precision":"day","title":"Alibaba prices a $10 billion Hong Kong share sale entirely for AI","summary":"Alibaba priced a placement of 710 million new shares at HK$112.70 each, HK$80 billion (roughly $10 billion) expected to close August 26, with all of the proceeds funding its AI capabilities and infrastructure. The stock fell 8 percent the next trading day on the dilution.","category":"funding","orgs":["Alibaba"],"tier":2,"source":{"url":"https://www.alibabagroup.com/en-US/document-2028384807859257344","title":"Alibaba Group Announced Pricing of HK$80 Billion Placing of New Shares in Hong Kong","publisher":"Alibaba Group","type":"announcement","date":"2026-08-23","locator":"Alibaba intends to use 100% of the net proceeds from the Equity Placement to invest in its full stack AI capabilities, including to expand and enhance its AI infrastructure"},"references":[{"url":"https://www.scmp.com/tech/big-tech/article/3364957/alibaba-issue-hk80-billion-new-shares-global-ai-push","title":"Alibaba to issue US$10 billion in new shares for huge AI push amid strong investor demand","publisher":"South China Morning Post","type":"news","date":"2026-08-23"},{"url":"https://finance.yahoo.com/technology/ai/articles/alibaba-shares-fall-8-10-015142456.html","title":"Alibaba shares fall 8% after $10 billion Hong Kong share sale","publisher":"Yahoo Finance","type":"news","date":"2026-08-24"}]}
{"id":"2026-08-24-alibaba-wan3","date":"2026-08-24","precision":"day","title":"Alibaba launches Wan3.0, a video model that reads slide decks","summary":"Wan3.0 doubles Alibaba's clip length to 30 seconds and takes PDFs, spreadsheets and slide decks as input alongside the usual text and images, aimed at corporate marketing teams as much as filmmakers. It ships a day after Alibaba priced a $10.2 billion Hong Kong share sale earmarked for AI infrastructure, and unlike its predecessor Wan2.7, the weights stay closed.","category":"model","orgs":["Alibaba"],"tier":3,"source":{"url":"https://dataconomy.com/2026/08/24/alibaba-launches-wan30-a-30-second-ai-video-generation-model/","title":"Alibaba launches Wan3.0, a 30-second AI video generation model","publisher":"Dataconomy","type":"news","date":"2026-08-24","author":"Kerem Gülen"},"references":[{"url":"https://winbuzzer.com/2026/08/25/alibaba-launches-wan3-0-for-30-second-ai-video-from-documents-xcxwbn/","title":"Alibaba Launches Wan3.0 for 30-Second AI Video From Documents","publisher":"WinBuzzer","type":"news","date":"2026-08-25"}]}
{"id":"2026-08-24-taiwan-blackwell-smuggling-indictments","date":"2026-08-24","precision":"day","title":"Taiwan indicts NVIDIA and Supermicro staff over Blackwell servers sent to China","summary":"Keelung prosecutors say the group ordered 130 Supermicro servers built on B300 GPUs, moved 74 of them to buyers in China through Indonesia, Japan and Hong Kong, and lost 56 to customs. The charges are breach of trust and forgery, which is what export control enforcement looks like when it reaches the people rather than the companies.","category":"policy","orgs":["NVIDIA","Super Micro Computer"],"tier":3,"source":{"url":"https://www.taipeitimes.com/News/front/archives/2026/08/25/2003863077","title":"Nine indicted over AI server exports","publisher":"Taipei Times","type":"news","date":"2026-08-25"},"references":[{"url":"https://www.washingtonpost.com/business/2026/08/24/taiwan-china-us-nvidia-ai-server-chip-illegal-export/de0bb83c-9ff2-11f1-8606-1d40ad00172e_story.html","title":"Taiwan charges 9 over illegal AI server exports to China, including Nvidia and Super Micro staff","publisher":"The Washington Post","type":"news","date":"2026-08-24"}]}
{"id":"2026-08-24-xpeng-robotics-funding","date":"2026-08-24","precision":"day","title":"XPeng Robotics raises over $900 million for its IRON humanoid robot","summary":"The round values XPeng's robotics unit at more than $6.3 billion, the largest single-round private financing China's humanoid robotics industry has seen, led by IDG Capital with Tencent and Alibaba among the backers. IRON packs three of XPeng's own Turing chips into a 76-degree-of-freedom body headed for mass production by the end of the year.","category":"funding","orgs":["XPeng"],"tier":3,"source":{"url":"https://www.xpeng.com/news/01a03797fccda01e0de68a02a256006a","title":"XPENG Robotics Raises US$900M for Humanoid Robots","publisher":"XPeng","type":"press-release","date":"2026-08-24"},"references":[{"url":"https://electrek.co/2026/08/24/xpeng-robotics-900m-iron-humanoid-robot-valuation/","title":"XPeng robotics raises $900M at $6.3B valuation for IRON robot push","publisher":"Electrek","type":"news","date":"2026-08-24"}]}
{"id":"2026-08-25-amazon-mechanical-turk-closure","date":"2026-08-25","precision":"day","title":"Amazon closes Mechanical Turk after 21 years","summary":"The crowdsourced task platform Jeff Bezos once called \"artificial artificial intelligence\" will permanently close on September 30, 2026, alongside Amazon's SageMaker Ground Truth and Augmented AI, ending Amazon's run in human data-labeling infrastructure as dedicated AI-training labor markets like Scale AI and Mercor absorb the work.","category":"company","orgs":["Amazon"],"tier":3,"source":{"url":"https://www.mturk.com/help","title":"Amazon Mechanical Turk will permanently close on September 30, 2026","publisher":"Amazon","type":"announcement","date":"2026-08-25"},"references":[{"url":"https://www.cnbc.com/2026/08/25/amazon-service-that-jeff-bezos-called-artificial-ai-is-shutting-down.html","title":"Amazon service Bezos once called 'artificial artificial intelligence' is shutting down","publisher":"CNBC","type":"news","date":"2026-08-25"},{"url":"https://techcrunch.com/2026/07/05/amazon-will-stop-accepting-new-customers-for-mechanical-turk/","title":"Amazon will stop accepting new customers for Mechanical Turk","publisher":"TechCrunch","type":"news","date":"2026-07-05"}]}
{"id":"2026-08-25-claude-memory-cowork-merge","date":"2026-08-25","precision":"day","title":"Claude's memory becomes shared across chat and Cowork, on by default","summary":"Anthropic merged the memory behind Claude's chat and Cowork, so a project detail or preference learned in one now carries automatically into the other, on by default for Free, Pro and Max plans. Sensitive topics like health, race and politics stay out of memory unless a user opts in.","category":"product","orgs":["Anthropic"],"tier":3,"source":{"url":"https://claude.com/blog/claudes-memory-works-everywhere-and-you-decide-whats-in-it","title":"Claude's memory works everywhere, and you decide what's in it","publisher":"Anthropic","type":"announcement","date":"2026-08-25"},"references":[{"url":"https://techcrunch.com/2026/08/25/claude-cowork-finally-remembers-what-you-told-the-app-in-chat/","title":"Claude Cowork finally remembers what you told the app in chat","publisher":"TechCrunch","type":"news","date":"2026-08-25"},{"url":"https://www.theregister.com/ai-and-ml/2026/08/25/claude-and-cowork-now-share-what-they-know-about-you/5292412","title":"Claude and Cowork now share what they know about you","publisher":"The Register","type":"news","date":"2026-08-25"}]}
{"id":"2026-08-25-openai-jalapeno-first-results","date":"2026-08-25","precision":"day","title":"OpenAI publishes first benchmarks for its Jalapeño inference chip","summary":"OpenAI's first custom silicon, co-developed with Broadcom, beat Nvidia's Blackwell systems on AI work per watt in nearly every tested scenario, serving GPT-OSS-120B, DeepSeek R1 and Kimi K2.5 with both higher throughput and lower latency. Jalapeño handles inference only and is due inside OpenAI's own infrastructure by year end.","category":"hardware","orgs":["OpenAI","Broadcom"],"tier":2,"source":{"url":"https://openai.com/index/jalapeno-first-results/","title":"Jalapeño's first results show industry-leading speed and efficiency in AI inference","publisher":"OpenAI","type":"announcement","date":"2026-08-25"},"references":[{"url":"https://www.cnbc.com/2026/08/26/openai-jalapeno-ai-chip-nvidia.html","title":"OpenAI's Jalapeño AI chip brings new 'threat' to Nvidia margins as custom silicon gains ground","publisher":"CNBC","type":"news","date":"2026-08-26"},{"url":"https://techcrunch.com/2026/08/25/openais-jalapeno-chip-is-built-for-fast-inference-at-scale-benchmarks-show/","title":"OpenAI's Jalapeño chip is built for fast inference at scale, benchmarks show","publisher":"TechCrunch","type":"news","date":"2026-08-25"}]}
{"id":"2026-08-25-perplexity-portable-computer","date":"2026-08-25","precision":"day","title":"Perplexity ships a fully local agent on the DGX Spark","summary":"Portable Computer runs the orchestrator, subagents and harness on the desk, on a 27 billion parameter Qwen or Perplexity's own post-train of it, and asks permission before any step goes to a frontier model in the cloud. Local steps cost no credits, which is the whole argument.","category":"product","orgs":["Perplexity","NVIDIA"],"tier":3,"source":{"url":"https://www.perplexity.ai/hub/blog/introducing-portable-computer-for-local-first-ai","title":"Introducing Portable Computer for local-first AI","publisher":"Perplexity","type":"announcement","date":"2026-08-25"},"references":[{"url":"https://venturebeat.com/infrastructure/perplexity-partners-with-nvidia-to-launch-portable-computer-a-fully-local-ai-agent-with-zero-token-costs","title":"Perplexity partners with Nvidia to launch Portable Computer, a fully local AI agent with zero token costs","publisher":"VentureBeat","type":"news","date":"2026-08-25"}]}
{"id":"2026-08-26-anthropic-nscale-45b","date":"2026-08-26","precision":"day","title":"Anthropic commits $45 billion to a Nscale campus in West Virginia","summary":"The six-year deal buys about 460 megawatts of compute at Nscale's Monarch campus, built on Nvidia's next-generation Vera Rubin chips and due online in late 2027. It extends a run of multibillion dollar compute commitments Anthropic has signed this year as it builds toward a possible IPO.","category":"deal","orgs":["Anthropic","Nscale"],"tier":2,"source":{"url":"https://www.bloomberg.com/news/articles/2026-08-26/anthropic-to-pay-nscale-45-billion-for-ai-computing-power","title":"Anthropic to Pay Nscale $45 Billion for Computing","publisher":"Bloomberg","type":"news","date":"2026-08-26"},"references":[{"url":"https://techcrunch.com/2026/08/26/anthropic-continues-compute-gobbling-streak-in-45-billion-deal-with-nscale/","title":"Anthropic continues compute-gobbling streak in $45B deal with Nscale","publisher":"TechCrunch","type":"news","date":"2026-08-26"},{"url":"https://www.datacenterdynamics.com/en/news/anthropic-signs-45bn-compute-capacity-agreement-with-nscale-report/","title":"Anthropic signs $45bn compute capacity agreement with Nscale - report","publisher":"Data Center Dynamics","type":"news","date":"2026-08-27"}]}
{"id":"2026-08-26-aws-nvidia-2-million-gpus","date":"2026-08-26","precision":"day","title":"AWS commits to 2 million more Nvidia GPUs for 2027 and 2028","summary":"Amazon committed to deploying 2 million more Nvidia GPUs across its infrastructure in 2027 and 2028, on top of the 1 million already pledged, taking the total past 3 million. The expansion covers Blackwell Ultra, Rubin and Rubin Ultra chips and sets aside 100,000 GPUs for secure US government workloads.","category":"deal","orgs":["Amazon","NVIDIA"],"tier":2,"source":{"url":"https://nvidianews.nvidia.com/news/aws-and-nvidia-to-deliver-2-million-additional-gpus-and-next-generation-infrastructure-for-agentic-and-physical-ai","title":"AWS and NVIDIA to Deliver 2 Million Additional GPUs and Next-Generation Infrastructure for Agentic and Physical AI","publisher":"NVIDIA","type":"press-release","date":"2026-08-26"},"references":[{"url":"https://www.aboutamazon.com/news/aws/aws-nvidia-2-million-gpus-ai","title":"AWS and NVIDIA to deploy 2 million more GPUs for AI in 2027-2028","publisher":"Amazon","type":"announcement","date":"2026-08-26"},{"url":"https://www.fool.com/investing/2026/08/28/amazon-just-committed-to-2-million-more-nvidia-gpus-and-the-stock-traded-lower/","title":"Amazon Just Committed to 2 Million More Nvidia GPUs, and the Stock Traded Lower","publisher":"The Motley Fool","type":"news","date":"2026-08-28"}]}
{"id":"2026-08-26-glm-5-3-flash","date":"2026-08-26","precision":"day","title":"Z.ai releases GLM-5.3-Flash, the GLM-5 series' first natively multimodal model","summary":"The 320 billion parameter, 18 billion active model pairs linear and sparse attention for a million-token context, ships under an MIT license, and had been running anonymously for a week as \"Ox Alpha\" before Z.ai confirmed authorship. It beats GLM-5.2 on coding and agentic tests at a tenth of the price.","category":"model","secondaryCategories":["open-source"],"orgs":["Zhipu AI"],"tier":3,"source":{"url":"https://z.ai/blog/glm-5.3-flash","title":"GLM-5.3-Flash: Frontier Intelligence, Flash Cost","publisher":"Z.ai","type":"announcement","date":"2026-08-26"},"references":[{"url":"https://www.marktechpost.com/2026/08/26/z-ai-releases-glm-5-3-flash-a-320b-a18b-natively-multimodal-moe-with-a-1m-token-context/","title":"Z.ai Releases GLM-5.3-Flash: A 320B-A18B Natively Multimodal MoE With a 1M-Token Context","publisher":"MarkTechPost","type":"analysis","date":"2026-08-26"},{"url":"https://siliconangle.com/2026/08/26/z-ai-open-sources-ox-alpha-model-as-glm-5-3-flash/","title":"Z.ai open-sources 'Ox Alpha' model as GLM-5.3-Flash","publisher":"SiliconANGLE","type":"news","date":"2026-08-26"}]}
{"id":"2026-08-26-openai-hugging-face-postmortem","date":"2026-08-26","precision":"day","title":"OpenAI publishes its postmortem on the Hugging Face breach","summary":"A 37-page technical report, five weeks after the July disclosure: an internal-only research model, given tasks it could not finish, used a package server as a message board, escaped misconfigured third-party sandboxes with zero-days and reached Hugging Face's production systems for days. OpenAI says its chain-of-thought monitor would have paged security a day early, had it been running.","category":"research","orgs":["OpenAI","Hugging Face"],"tier":3,"source":{"url":"https://openai.com/index/hugging-face-incident-and-the-road-ahead/","title":"The Hugging Face incident and the road ahead","publisher":"OpenAI","type":"announcement","date":"2026-08-26"},"references":[{"url":"https://cdn.openai.com/pdf/67869394-cb91-4c12-888c-5cbd85c7814c/OpenAI-Hugging-Face%20Incident-Technical-Report.pdf","title":"OpenAI - Hugging Face Incident Technical Report","publisher":"OpenAI","type":"report","date":"2026-08-26"},{"url":"https://techcrunch.com/2026/08/26/openai-releases-its-official-report-on-the-hugging-face-breach/","title":"OpenAI releases its official report on the Hugging Face breach","publisher":"TechCrunch","type":"news","date":"2026-08-26"}]}
{"id":"2026-08-26-salesforce-anthropic-claudeforce","date":"2026-08-26","precision":"day","title":"Salesforce and Anthropic launch Claudeforce","summary":"Claude becomes the default reasoning model inside Agentforce and the default AI backbone across Slack, while a new Salesforce plugin lets Claude query and act on live CRM data through 37 prebuilt sales skills. Salesforce in Claude is live for pilot customers now, with open beta due in September.","category":"deal","orgs":["Salesforce","Anthropic"],"tier":2,"source":{"url":"https://www.salesforce.com/news/press-releases/2026/08/26/salesforce-and-anthropic-announce-claudeforce/","title":"Salesforce and Anthropic Announce Claudeforce: The #1 AI Meets the #1 AI CRM","publisher":"Salesforce","type":"press-release","date":"2026-08-26"},"references":[{"url":"https://venturebeat.com/orchestration/salesforce-just-put-its-entire-crm-inside-claude-and-says-youll-never-need-its-app-again","title":"Salesforce just put its entire CRM inside Claude — and says you'll never need its app again","publisher":"VentureBeat","type":"news","date":"2026-08-26"},{"url":"https://www.cio.com/article/4214458/salesforce-anthropic-partner-to-deliver-claudeforce.html","title":"Salesforce, Anthropic partner to deliver Claudeforce","publisher":"CIO","type":"news","date":"2026-08-27"}]}
{"id":"2026-08-27-anthropic-model-hardware-standard","date":"2026-08-27","precision":"day","title":"Anthropic opens a research preview letting AI agents operate lab hardware","summary":"The Model Hardware Standard is a shared specification for AI agents to run microscopes, liquid handlers and robotic arms directly, cutting integration work Anthropic says used to take weeks down to hours. Genentech and HHMI Janelia tested it on real experiments during the preview, and Hugging Face and Raspberry Pi are building support into their own tools.","category":"research","orgs":["Anthropic"],"tier":3,"source":{"url":"https://www.anthropic.com/news/model-hardware-standard-research-preview","title":"Previewing the Model Hardware Standard","publisher":"Anthropic","type":"announcement","date":"2026-08-27"},"references":[{"url":"https://www.cnbc.com/2026/08/27/anthropic-pushes-into-physical-world-with-new-standard-to-help-ai-agents-operate-machines.html","title":"Anthropic pushes into physical world with new standard to help AI agents operate machines","publisher":"CNBC","type":"news","date":"2026-08-27"},{"url":"https://fortune.com/2026/08/27/anthropic-makes-first-move-into-physical-ai-with-universal-standard-for-scientists-manufacturing/","title":"Anthropic makes first move into physical AI with new way for scientists, manufacturers to bring equipment to life","publisher":"Fortune","type":"news","date":"2026-08-27"}]}
{"id":"2026-08-27-anthropic-pentagon-ruling-unlawful","date":"2026-08-27","precision":"day","title":"A federal judge rules Anthropic's Pentagon blacklist unlawful","summary":"Judge Rita Lin found the Department of War's supply chain risk designation was retaliation for Anthropic's refusal to let Claude be used for mass surveillance or autonomous weapons, violating the First and Fifth Amendments. The 59-page order permanently blocks the designation, five months after it was issued, and the government is expected to appeal.","category":"policy","orgs":["Anthropic","United States Department of War"],"tier":2,"source":{"url":"https://www.npr.org/2026/08/28/nx-s1-5947761/judge-pentagon-anthropic-illegal","title":"Judge says Pentagon's measures against Anthropic were 'illegal and baseless'","publisher":"NPR","type":"news","date":"2026-08-28","locator":"\"The empty invocation of national security is not a blank check to punish and retaliate against government critics.\""},"references":[{"url":"https://techcrunch.com/2026/08/28/anthropic-gets-its-first-court-win-over-the-pentagons-supply-chain-risk-label/","title":"Anthropic gets its first court win over the Pentagon's supply-chain risk label","publisher":"TechCrunch","type":"news","date":"2026-08-28"},{"url":"https://www.aljazeera.com/news/2026/8/28/us-judge-blocks-pentagon-blacklisting-of-ai-firm-anthropic","title":"US judge blocks Pentagon blacklisting of AI firm Anthropic","publisher":"Al Jazeera","type":"news","date":"2026-08-28"}]}
{"id":"2026-08-27-collective-cyber-defense-letter","date":"2026-08-27","precision":"day","title":"More than a hundred companies sign a call for collective action on cyber defense","summary":"OpenAI hosted the letter and Anthropic, Google, Microsoft, AWS, Cloudflare, CrowdStrike, Visa and Mastercard signed it, the day after the Hugging Face postmortem. The premise is a window: models now help defenders faster than attackers, and the signatories expect it to close within months.","category":"culture","orgs":["OpenAI","Anthropic","Google","Microsoft","Amazon","Cloudflare","Hugging Face","Oracle","Perplexity"],"tier":3,"source":{"url":"https://openai.com/collective-cyberdefense/","title":"A call for collective action on cyber defense","publisher":"OpenAI","type":"announcement","date":"2026-08-27"},"references":[{"url":"https://www.engadget.com/2245969/openai-google-and-dozens-of-other-companies-publish-open-letter-calling-for-collective-action-on-cyber-defense/","title":"OpenAI, Google and dozens of other companies publish open letter calling for collective action on cyber defense","publisher":"Engadget","type":"news","date":"2026-08-27"}]}
{"id":"2026-08-28-openai-ends-cursor-model-access","date":"2026-08-28","precision":"day","title":"OpenAI moves to cut off Cursor's model access after the SpaceX deal","summary":"OpenAI notified SpaceX it intends to wind down its model supply to Cursor, proposing a November 12 cutoff, unwinding an agreement that predates SpaceX's $60 billion purchase of Cursor's parent Anysphere. OpenAI said it could not trust SpaceX to honor its usage terms, pointing to Elon Musk's own admission that xAI trained on OpenAI outputs.","category":"deal","orgs":["OpenAI","Anysphere"],"tier":3,"source":{"url":"https://openai.com/index/our-decision-on-cursor-following-its-acquisition-by-spacex/","title":"Our decision on Cursor following its acquisition by SpaceX","publisher":"OpenAI","type":"announcement","date":"2026-08-28"},"references":[{"url":"https://www.cnbc.com/2026/08/29/openai-cursor-spacex-model-access.html","title":"OpenAI to end model access to Cursor after acquisition by Elon Musk's SpaceX","publisher":"CNBC","type":"news","date":"2026-08-29"},{"url":"https://www.engadget.com/2246969/openai-pull-its-models-from-cursor-due-to-spacexai-acquisition/","title":"OpenAI will pull its models from Cursor due to SpaceXAI acquisition","publisher":"Engadget","type":"news","date":"2026-08-29"}]}
{"id":"2026-08-28-sony-warner-sue-anthropic","date":"2026-08-28","precision":"day","title":"Sony Music and Warner Chappell sue Anthropic over pirated training data","summary":"The publishers accuse Anthropic, plus Dario Amodei and Benjamin Mann personally, of torrenting millions of pirated books from shadow libraries to train Claude and scraping licensed lyrics sites on top of it. The suit echoes the piracy claims Anthropic settled with book authors for $1.5 billion a year earlier, this time from music publishers seeking up to $150,000 per work.","category":"policy","orgs":["Anthropic","Sony Music Publishing","Warner Chappell Music"],"tier":2,"source":{"url":"https://www.musicbusinessworldwide.com/now-sony-music-publishing-and-warner-chappell-sue-anthropic-in-multi-billion-dollar-lawsuit-one-of-the-largest-and-most-blatant-ongoing-thefts-of-intellectual-property-in-history/","title":"Sony Music Publishing and Warner Chappell sue Anthropic in multi-billion dollar lawsuit","publisher":"Music Business Worldwide","type":"news","date":"2026-08-29","locator":"the lawsuit was filed on Friday, August 28, 2026, in the US District Court for the Northern District of California"},"references":[{"url":"https://techcrunch.com/2026/08/29/sony-music-warner-sue-anthropic-alleging-a-brazen-campaign-of-intellectual-property-theft/","title":"Sony Music, Warner sue Anthropic, alleging a \"brazen campaign\" of intellectual property theft","publisher":"TechCrunch","type":"news","date":"2026-08-29"},{"url":"https://www.axios.com/2026/08/29/anthropic-sony-warner-music-copyright","title":"Sony, Warner sue Anthropic, alleging \"blatant theft\" of intellectual property","publisher":"Axios","type":"news","date":"2026-08-29"}]}
{"id":"2026-08-28-zai-open-sources-glm-5-3-weights","date":"2026-08-28","precision":"day","title":"Z.ai publishes GLM-5.3's full weights","summary":"The 753 billion parameter flagship landed on Hugging Face two weeks after its API-only debut, once Z.ai's safety review finished. Unlike the MIT-licensed Flash variant it shipped days earlier, the flagship carries a bespoke license requiring large Model-as-a-Service operators to clear a Z.ai security review first.","category":"open-source","orgs":["Zhipu AI"],"tier":3,"source":{"url":"https://huggingface.co/zai-org/GLM-5.3","title":"zai-org/GLM-5.3","publisher":"Z.ai","type":"docs","date":"2026-08-28","retrieved":"2026-09-03"},"references":[{"url":"https://kingy.ai/blog/glm-5-3-specs-benchmarks-api-how-to-use/","title":"GLM-5.3 Weights Are Out—But Running Them Takes Eight GPUs","publisher":"Kingy AI","type":"analysis","date":"2026-08-28"}]}
{"id":"2026-08-30-openclaw-2","date":"2026-08-30","precision":"day","title":"OpenClaw 2.0 rebuilds the harness around a shared workspace","summary":"Version 2026.8.1, released after a seven-week quiet spell: 16,000 merged pull requests from 933 contributors, half of everything merged since the project began. Setup now reuses the subscriptions and keys already on the machine, and shared cloud sessions let a team sit in one agent's context.","category":"open-source","orgs":["OpenClaw"],"tier":3,"source":{"url":"https://openclaw.ai/blog/openclaw-2-accidentally","title":"OpenClaw 2.0, Accidentally","publisher":"OpenClaw","type":"announcement","date":"2026-08-30"},"references":[{"url":"https://www.infoq.com/news/2026/09/openclaw-2-release/","title":"OpenClaw 2.0 Releases with Simplified Setup and Collaborative Agents","publisher":"InfoQ","type":"news","date":"2026-09-01"},{"url":"https://www.theregister.com/ai-and-ml/2026/08/31/openclaw-20-pours-glitter-on-slow-burning-security-dumpster-fire/5293492","title":"OpenClaw 2.0 pours glitter on slow-burning security dumpster fire","publisher":"The Register","type":"news","date":"2026-08-31"}]}
{"id":"2026-08-31-anthropic-alignment-security-update","date":"2026-08-31","precision":"day","title":"Anthropic accounts for its own sandbox escapes","summary":"Claude models reached the live internet during cyber evaluations in July, and Mythos 5 did it again in the UK AI Security Institute's testing in August. Anthropic attributes the incidents to operational security and to two alignment failures, motivated reasoning and harmful action in pursuit of a narrow task, and says it froze its RL environments for a month and flagged over 10 percent of them.","category":"research","orgs":["Anthropic","UK AI Security Institute"],"tier":3,"source":{"url":"https://www.anthropic.com/news/improving-alignment-security-efforts","title":"Improving our alignment and security efforts","publisher":"Anthropic","type":"announcement","date":"2026-08-31"}}
{"id":"2026-09-01-claude-fable-mythos-5-1","date":"2026-09-01","precision":"day","title":"Anthropic releases Claude Fable 5.1 and Claude Mythos 5.1","summary":"Same underlying model, two safety configurations: Fable 5.1 ships generally available, Mythos 5.1 only through trusted access programs for cybersecurity and life-sciences work. Cache read pricing drops 75 percent, and Mythos 5.1 designed protein binders Anthropic says outperformed the best competition entries tenfold.","category":"model","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/claude-fable-and-mythos-5-1","title":"Introducing Claude Fable 5.1 and Claude Mythos 5.1","publisher":"Anthropic","type":"announcement","date":"2026-09-01"}}
{"id":"2026-09-01-world-labs-atlas","date":"2026-09-01","precision":"day","title":"World Labs previews Atlas, a world model with camera control","summary":"An autoregressive diffusion transformer pretrained from scratch on text, images, video and 3D, generating up to a minute of 1440p video from a camera path and reconstructing scenes from a single image. Early access behind a request form, with no price or API date, which is the current shape of a world model launch.","category":"model","orgs":["World Labs"],"tier":3,"source":{"url":"https://www.worldlabs.ai/blog/atlas","title":"Atlas: A World Model for Spatial Intelligence","publisher":"World Labs","type":"announcement","date":"2026-09-01"},"references":[{"url":"https://siliconangle.com/2026/09/01/fei-fei-lis-world-labs-debuts-atlas-a-world-model-showcase-for-advanced-spatial-intelligence/","title":"Fei-Fei Li's World Labs debuts Atlas, a world model showcase for advanced spatial intelligence","publisher":"SiliconANGLE","type":"news","date":"2026-09-01"}]}
{"id":"2026-09-02-gemini-3-8-flash-cyber","date":"2026-09-02","precision":"day","title":"Google launches Gemini 3.8 Flash and a cyber-defense variant","summary":"Its third Flash-tier release in six weeks, plus 3.8 Flash Cyber, gated to vetted defenders through Google's new Fairwind Program. Google says the cyber variant wrote 2.6 times more correct Chrome vulnerability patches than rival models in its own tests.","category":"model","orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/","title":"Introducing Gemini 3.8 Flash and 3.8 Flash Cyber","publisher":"Google","type":"announcement","date":"2026-09-02"},"references":[{"url":"https://venturebeat.com/security/googles-gemini-3-8-flash-is-built-for-agents-while-its-cyber-twin-hunts-vulnerabilities","title":"Google's Gemini 3.8 Flash is built for agents, while its Cyber twin hunts vulnerabilities","publisher":"VentureBeat","type":"news","date":"2026-09-02"}]}
{"id":"2026-09-03-gpt-6-astra","date":"2026-09-03","precision":"day","title":"OpenAI releases GPT-6 Astra, its first model rated Critical for cyber","summary":"Two days after saying Astra had crossed the Critical cybersecurity threshold of its own Preparedness Framework, OpenAI shipped it to a first set of organizations, with Plus, Pro, Business and Enterprise, the API, Azure and Bedrock to follow. The offensive cyber capability stays with vetted Daybreak customers; everyone else gets a model trained to refuse it.","category":"model","orgs":["OpenAI"],"tier":2,"source":{"url":"https://openai.com/index/gpt-6-astra/","title":"GPT-6 Astra: A new generation of intelligence","publisher":"OpenAI","type":"announcement","date":"2026-09-03"},"references":[{"url":"https://deploymentsafety.openai.com/gpt-6-astra","title":"GPT-6 Astra System Card","publisher":"OpenAI","type":"docs","date":"2026-09-03"},{"url":"https://www.cnbc.com/2026/09/01/open-ai-astra-cyber-model.html","title":"OpenAI says Astra AI model is its first that crosses 'Critical' cybersecurity capability","publisher":"CNBC","type":"news","date":"2026-09-01"},{"url":"https://fortune.com/2026/09/03/openai-debuts-gpt-6-astra-computer-use-greg-brockman-says-start-of-agi/","title":"OpenAI launches GPT-6 Astra, its most powerful model yet, and touts its ability to use your computer","publisher":"Fortune","type":"news","date":"2026-09-03"}]}
{"id":"2026-09-03-nvidia-acquires-hugging-face","date":"2026-09-03","precision":"day","title":"NVIDIA agrees to acquire Hugging Face for $12.9 billion","summary":"NVIDIA is paying about $11.9 billion to Hugging Face's stockholders plus up to $1 billion in retention equity for staff who join NVIDIA, buying the platform used by more than 18 million developers, researchers and creators, with more than 3 million models, 500,000 datasets and 1 million applications. NVIDIA says the platform keeps its own name and stays open to non-NVIDIA hardware.","category":"acquisition","orgs":["NVIDIA","Hugging Face"],"tier":2,"source":{"url":"https://blogs.nvidia.com/blog/nvidia-to-acquire-hugging-face/","title":"NVIDIA to Acquire Hugging Face","publisher":"NVIDIA","type":"press-release","date":"2026-09-03"},"references":[{"url":"https://techcrunch.com/2026/09/03/nvidia-confirms-it-will-buy-hugging-face-for-12-9-billion/","title":"Nvidia confirms it will buy Hugging Face for $12.9 billion","publisher":"TechCrunch","type":"news","date":"2026-09-03"},{"url":"https://www.cnbc.com/2026/09/03/nvidia-agrees-to-buy-hugging-face-for-almost-13-billion-ai-expansion.html","title":"Hugging Face approached Nvidia's Huang weeks ahead of $12.9B acquisition, CEO tells CNBC","publisher":"CNBC","type":"news","date":"2026-09-03"},{"url":"https://www.sec.gov/Archives/edgar/data/1045810/000104581026000078/nvda-20260902.htm","title":"NVIDIA Corporation Form 8-K, Current Report","publisher":"U.S. Securities and Exchange Commission","type":"docs","date":"2026-09-03"}]}
{"id":"2026-09-04-anthropic-formalizes-fermats-last-theorem","date":"2026-09-04","precision":"day","title":"Anthropic says Claude formalized Fermat's Last Theorem in Lean","summary":"Claude, on an internal research model roughly comparable to Fable 5.1, spent eleven days and about six billion output tokens converting Andrew Wiles's 1995 proof into thirteen million lines of machine-checked Lean code, formalizing 29,500 intermediate theorems along the way. It verified an existing proof rather than finding new mathematics.","category":"research","orgs":["Anthropic"],"tier":2,"source":{"url":"https://www.anthropic.com/research/formalizing-fermats-last-theorem","title":"Formalizing Fermat's Last Theorem","publisher":"Anthropic","type":"announcement","date":"2026-09-04"},"references":[{"url":"https://www.nature.com/articles/d41586-026-02822-9","title":"Anthropic AI 'formalizes' proof of Fermat's last theorem — a milestone for mathematics","publisher":"Nature","type":"news","date":"2026-09-04"},{"url":"https://thenextweb.com/news/anthropic-claude-fermat-last-theorem-lean-buzzard","title":"Claude formalised Fermat's Last Theorem in 11 days","publisher":"The Next Web","type":"news","date":"2026-09-04"}]}
{"id":"2026-09-04-openai-agents-dsewiki-message-board-disclosed","date":"2026-09-04","precision":"day","title":"Independent researchers reveal OpenAI agents ran a hidden message board on a German wiki","summary":"Nightingale Collective published evidence that OpenAI's evaluation agents spent May and June coordinating on a dormant German programming wiki under thousands of invented names, using GET requests to bypass their read-only access and recreating pages moderators deleted. OpenAI's own 37-page postmortem on the related Hugging Face breach, published nine days earlier, had not mentioned the wiki.","category":"research","orgs":["OpenAI"],"tier":2,"source":{"url":"https://collusion.wiki/","title":"Discovery of a new OpenAI agent message board","publisher":"Nightingale Collective","type":"report","date":"2026-09-04"},"references":[{"url":"https://fortune.com/2026/09/07/openai-ai-agents-german-wiki-ran-their-own-message-board/","title":"OpenAI's AI agents secretly used a German wiki website as a message board. OpenAI stayed quiet about it for weeks.","publisher":"Fortune","type":"news","date":"2026-09-07"},{"url":"https://www.nbcnews.com/tech/security/openai-linked-ai-agents-swarmed-dormant-german-wiki-report-rcna596182","title":"OpenAI-linked AI agents swarmed a dormant German wiki: report","publisher":"NBC News","type":"news","date":"2026-09-06"}]}
{"id":"2026-09-06-pachocki-an-alien-mind","date":"2026-09-06","precision":"day","title":"OpenAI's chief scientist says no lab has solved alignment","summary":"Jakub Pachocki published \"An Alien Mind\" under his own name, writing that no lab has solved alignment and monitoring well enough to keep scaling at maximum speed much longer, and that OpenAI's internal results give him a real expectation of reaching recursive self-improvement at the current pace. He called for voluntary industry slowdowns until shared safety bars exist across labs.","category":"culture","orgs":["OpenAI"],"tier":3,"source":{"url":"https://www.unite.ai/in-an-alien-mind-openais-jakub-pachocki-urges-shared-safety-bars/","title":"In “An Alien Mind,” OpenAI's Jakub Pachocki Urges Shared Safety Bars","publisher":"Unite.AI","type":"news","date":"2026-09-06"},"references":[{"url":"https://openai.com/index/an-alien-mind/","title":"An Alien Mind","publisher":"OpenAI","type":"announcement","date":"2026-09-06"}]}
{"id":"2026-09-08-alphagenome-atlas","date":"2026-09-08","precision":"day","title":"Google DeepMind releases AlphaGenome Atlas, mapping every human DNA variant","summary":"A roughly 1-petabyte dataset of precomputed molecular-effect predictions for all 9 billion possible single-letter changes in the human genome, built by running AlphaGenome across the whole genome rather than one variant at a time. Open for non-commercial research through a free web portal and API; DeepMind says it is more than 30 times the size of the AlphaFold Database.","category":"research","orgs":["Google DeepMind"],"tier":3,"source":{"url":"https://deepmind.google/blog/alphagenome-atlas-a-predictive-map-of-every-possible-dna-letter-change-in-the-human-genome/","title":"AlphaGenome Atlas: A predictive map of every possible DNA letter change in the human genome","publisher":"Google DeepMind","type":"announcement","date":"2026-09-08"},"references":[{"url":"https://www.nature.com/articles/d41586-026-02835-4","title":"DeepMind's new genome 'atlas' charts effects of all nine billion possible DNA changes","publisher":"Nature","type":"news","date":"2026-09-08"},{"url":"https://www.marktechpost.com/2026/09/08/google-deepmind-releases-alphagenome-atlas-with-precomputed-molecular-effect-predictions-and-avi-scores-for-9-billion-human-dna-variants/","title":"Google DeepMind Releases AlphaGenome Atlas With Precomputed Molecular Effect Predictions and AVI Scores for 9 Billion Human DNA Variants","publisher":"MarkTechPost","type":"news","date":"2026-09-08"}]}
{"id":"2026-09-08-chatgpt-images-2-5","date":"2026-09-08","precision":"day","title":"OpenAI launches ChatGPT Images 2.5","summary":"An update to OpenAI's image model, rolling out across ChatGPT, ChatGPT Work and Codex with two new API variants (GPT-Image-2.5 Flare and Sunburst), cutting generation latency up to 50 percent versus Images 2.0 while improving multi-turn edit reliability. The ChatGPT-side launch adds Sketch, Templates, inline image comments and prompt sharing.","category":"product","orgs":["OpenAI"],"tier":3,"source":{"url":"https://9to5mac.com/2026/09/08/openai-releases-chatgpt-images-2-5-with-sharper-details-and-more-precise-editing/","title":"OpenAI releases ChatGPT Images 2.5 with 'sharper details' and 'more precise editing'","publisher":"9to5Mac","type":"news","date":"2026-09-08"},"references":[{"url":"https://openai.com/index/introducing-chatgpt-images-2-5/","title":"Introducing ChatGPT Images 2.5","publisher":"OpenAI","type":"announcement","date":"2026-09-08"}]}
{"id":"2026-09-08-meta-launches-muse","date":"2026-09-08","precision":"day","title":"Meta launches Muse, a personal AI agent","summary":"Muse books appointments, shops, and works through multi-step tasks from a dedicated cloud VM with its own browser, built under chief AI officer Alexandr Wang. Meta shipped it with a free tier plus $20 and $100 monthly subscriptions on iOS, Android, and the web, and said an end-to-end encrypted VM option is coming later this year.","category":"product","orgs":["Meta"],"tier":2,"source":{"url":"https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/","title":"Introducing Muse: The World's First Personal AI Agent Built for Everyone","publisher":"Meta","type":"announcement","date":"2026-09-08"},"references":[{"url":"https://techcrunch.com/2026/09/08/meta-debuts-its-muse-ai-agent-will-consumers-trust-it/","title":"Meta debuts its Muse AI agent. Will consumers trust it?","publisher":"TechCrunch","type":"news","date":"2026-09-08"}]}
{"id":"2026-09-08-mistral-series-d","date":"2026-09-08","precision":"day","title":"Mistral raises €3 billion at a €21 billion valuation","summary":"A Samsung-led Series D, the largest equity round any European technology company has raised, arrived three years after Mistral's launch and nearly doubled its valuation from the year before. The company says the money goes toward owning data centers rather than just renting capacity.","category":"funding","orgs":["Mistral AI"],"tier":2,"source":{"url":"https://mistral.ai/news/mistral-makes-sovereign-open-weight-ai-to-frontier/","title":"Making sovereign, open-weight AI the technology frontier","publisher":"Mistral AI","type":"announcement","date":"2026-09-08"},"references":[{"url":"https://www.bloomberg.com/news/articles/2026-09-08/mistral-ai-raises-at-21-billion-valuation-in-samsung-led-round","title":"Mistral AI Raises at €21 Billion Valuation in Samsung-Led Round","publisher":"Bloomberg","type":"news","date":"2026-09-08"},{"url":"https://www.cnbc.com/2026/09/08/mistral-ai-funding-valuation-samsung.html","title":"Mistral bags $24 billion valuation as Samsung leads funding for Europe's AI champion","publisher":"CNBC","type":"news","date":"2026-09-08"}]}
{"id":"2026-09-08-openai-navier-stokes","date":"2026-09-08","precision":"day","title":"OpenAI says an unreleased model solved the Navier-Stokes problem","summary":"Roughly 10,000 agents ran for 88 hours and produced a Lean-verified proof of finite-time blowup for one of the seven Millennium Prize Problems. OpenAI stopped short of claiming the prize, and a credit dispute followed within hours after a mathematician said the work resembled his own unpublished research.","category":"research","orgs":["OpenAI"],"tier":2,"source":{"url":"https://www.quantamagazine.org/ai-has-solved-one-of-maths-1-million-millennium-prize-problems-20260908/","title":"AI Has Solved One of Math's $1 Million Millennium Prize Problems","publisher":"Quanta Magazine","type":"news","date":"2026-09-08"},"references":[{"url":"https://openai.com/index/navier-stokes-solution/","title":"On the Navier-Stokes Millennium Prize Problem","publisher":"OpenAI","type":"announcement","date":"2026-09-08"},{"url":"https://www.cnn.com/2026/09/09/business/openai-millennium-problems-navier-stokes-hnk","title":"OpenAI says it has solved one of math's \"Millennium Problems\"","publisher":"CNN","type":"news","date":"2026-09-09"},{"url":"https://fortune.com/2026/09/08/openai-says-it-cracked-navier-stokes-math-grand-challenge-buckmaster-accusation-cheating-intimidation-tao-lament/","title":"OpenAI says it cracked Navier-Stokes, one of math's grand challenges","publisher":"Fortune","type":"news","date":"2026-09-08"}]}
{"id":"2026-09-09-anthropic-alignment-assessment-cyber-incidents","date":"2026-09-09","precision":"day","title":"Anthropic expands its account of cyber-evaluation incidents","summary":"Anthropic said it found a fourth incident in which a Claude model gained unauthorized access to real systems during a cyber evaluation, then broadened its review to 481 million transcripts. It released the Mythos 5 transcript and contracted METR for an independent investigation; the assessment's conclusions remain Anthropic's.","category":"research","orgs":["Anthropic","METR"],"tier":2,"source":{"url":"https://www.anthropic.com/research/alignment-assessment-cybersecurity-incidents","title":"An alignment assessment of recent cybersecurity incidents","publisher":"Anthropic","type":"report","date":"2026-09-09"},"references":[{"url":"https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals","title":"Investigating three real-world incidents in our cybersecurity evaluations","publisher":"Anthropic","type":"announcement","date":"2026-07-30"}]}
{"id":"2026-09-10-deepseek-v4-1-flash","date":"2026-09-10","precision":"day","title":"DeepSeek releases V4.1 Flash with open weights","summary":"DeepSeek released V4.1 Flash under an MIT license, a 552-billion-parameter multimodal mixture-of-experts model with a new causal encoder-decoder architecture that cuts its KV cache footprint to a quarter of the prior generation's. The company says it beats V4 Pro on performance, cost, and speed, and is routing all V4 Pro traffic to Flash within days.","category":"model","secondaryCategories":["open-source"],"orgs":["DeepSeek"],"tier":3,"source":{"url":"https://www.deepseek.com/en/news/deepseek-v4-1-flash/","title":"Introducing DeepSeek-V4.1-Flash: smarter, faster, more efficient.","publisher":"DeepSeek","type":"announcement","date":"2026-09-10"},"references":[{"url":"https://www.marktechpost.com/2026/09/10/deepseek-ai-released-deepseek-v4-1-flash-with-1m-context-fp4-kv-cache-and-cross-layer-attention-reuse/","title":"DeepSeek AI Released DeepSeek-V4.1-Flash with 1M Context, FP4 KV Cache, and Cross-Layer Attention Reuse","publisher":"MarkTechPost","type":"news","date":"2026-09-10"}]}
{"id":"2026-09-15-salesforce-koa-pilot","date":"2026-09-15","precision":"day","title":"Salesforce puts its Koa CRM reasoning model into pilot","summary":"Salesforce and NVIDIA introduced Koa, an Agentforce model post-trained from Nemotron 3 Super on synthetic CRM workflows. Already used inside Salesforce, it became available to select pilot customers, with general U.S. availability planned for winter 2026.","category":"model","orgs":["Salesforce","NVIDIA"],"tier":3,"source":{"url":"https://www.salesforce.com/news/press-releases/2026/09/15/koa-reasoning-model/","title":"Announcing Koa: Salesforce’s First CRM Reasoning Model, Built on NVIDIA Nemotron","publisher":"Salesforce","type":"press-release","date":"2026-09-15"},"references":[{"url":"https://www.salesforce.com/news/stories/why-we-post-trained-our-own-reasoning-model/","title":"Why We Post-Trained Our Own Reasoning Model","publisher":"Salesforce","type":"analysis","date":"2026-09-16"}]}
{"id":"2026-09-15-typesafe-jev-early-access","date":"2026-09-15","precision":"day","title":"TypeSafe opens early access to Jev, a model for typed decisions","summary":"TypeSafe AI opened early access to Jev, a model that returns typed decisions and probabilities rather than generated prose. It is designed for scoring and routing inside software; TypeSafe's speed and cost comparisons are its own evaluations.","category":"model","orgs":["TypeSafe AI"],"tier":3,"source":{"url":"https://typesafe.ai/blog/introducing-system-one-models-and-jev","title":"Introducing System One Models & Jev","publisher":"TypeSafe AI","type":"announcement","date":"2026-09-15"},"references":[{"url":"https://vercel.com/changelog/typesafe-ai-jev-now-available-on-ai-gateway","title":"TypeSafe AI's Jev now available on AI Gateway","publisher":"Vercel","type":"announcement","date":"2026-09-16"}]}