Ilya Sutskever Leaves OpenAI as Superalignment Team Dissolves

A short, quiet post from the chief scientist

Ilya Sutskever posted on X at 7:00 PM on May 14, 2024 that he had decided to leave OpenAI after almost a decade, calling the company's trajectory "nothing short of miraculous" and naming Jakub Pachocki as the researcher who would now lead the work. He said he was moving to a project "very personally meaningful" to him, without naming it. [3]

Sam Altman replied on X the same day, calling Sutskever "one of the greatest minds of our generation." OpenAI named research director Jakub Pachocki, at the company since 2017 and a lead on GPT-4, as Sutskever's replacement as chief scientist. [2]

Atlas interpretation: The post came six months after Sutskever, then a board member, voted to remove Altman, a reversal that cost him his board seat while he stayed on staff. Altman had told reporters in March that there was "nothing to announce" about Sutskever's status, so the departure ended months of public ambiguity rather than arriving without warning. [2]

Three days later, a longer explanation

Jan Leike, who had co-led the superalignment team with Sutskever, posted on X on May 17, 2024 that his last day as "head of alignment, superalignment lead, and executive" had been the day before. He followed with a thread rather than a single statement. [4]

Leike wrote that he had "been disagreeing with OpenAI leadership about the company's core priorities for quite some time, until we finally reached a breaking point," and that his team had been "sailing against the wind," struggling for computing resources needed for its research. [4]

His most quoted line: "Building smarter-than-human machines is an inherently dangerous endeavor... But over the past years, safety culture and processes have taken a backseat to shiny products." He closed by telling OpenAI staff to "learn to feel the AGI" and act with matching gravitas. [4]

The team the departures left behind

OpenAI had announced the superalignment team in July 2023, committing 20% of the compute it had then secured toward the four-year problem of controlling AI systems smarter than their overseers. Less than a year later, a person familiar with the matter confirmed to CNBC that OpenAI had disbanded the team and was reassigning its remaining members to other research groups. [2]

OpenAI did not comment directly and pointed CNBC to a post from Altman saying he was sad to see Leike go and that the company had more work to do, plus a joint statement from Altman and Greg Brockman arguing OpenAI had "raised awareness of the risks and opportunities of AGI so that the world can better prepare for it." [2]