The superalignment leads leave OpenAI
On 14 May 2024 Ilya Sutskever announced that he was leaving OpenAI, and on 17 May Jan Leike wrote that the previous day had been his last as head of alignment and superalignment lead. Leike explained that he had long disagreed with leadership about the company's priorities until a breaking point was reached, and that the team had been short of compute.
Why it matters
Both leaders of a team promised a fifth of the compute less than a year earlier left within three days, and one of them said publicly that safety culture and processes had taken a backseat to 'shiny products'.
Leike's thread: much more effort should go to preparing for the next generations of models, on security, monitoring, preparedness, adversarial robustness, alignment and societal impact; 'Sometimes we were struggling for compute'; 'safety culture and processes have taken a backseat to shiny products'; OpenAI must become a safety-first AGI company. Sutskever says only that he is leaving after almost a decade and that he is confident OpenAI will build safe and beneficial AGI; he says nothing about safety or the team. The record does not claim that the team was disbanded, nor what share of compute it received: neither of the two posts read says so, and no company statement of a disbanding was found. The reply by Altman and Brockman was not read.