Sam Altman Says OpenAI Let Its AI Math Push Run Out of Control for One Week
JUST IN: Sam Altman admits OpenAI let its AI-generated math research get "out of control for one week” amid backlash from mathematicians.
Context
The head of OpenAI is defending a flood of AI generated math results that mathematicians say they cannot check fast enough, and says more are coming.
In recent weeks OpenAI has released more than 700 AI generated proofs, including a Navier Stokes result, and many mathematicians reacted with alarm. Sam Altman discussed the backlash on the Cheeky Pint podcast with Stripe cofounder John Collison, released on Thursday, Oct 29. Polymarket posted the quote on Saturday, and it spread widely.
Deep dive
On the Cheeky Pint podcast with John Collison, Sam Altman said OpenAI "did let it get out of control for one week on the math," and that this was intentional. According to the published transcript, he said the company had to push hard to learn how far teams of AI agents could go, because when it did the Navier Stokes proof it understood almost nothing about how to scale them: which model to use, how many copies, and how long to let them run.
Altman put the cost of the Navier Stokes work at around $10 million, "maybe a little more," and said later results cost far less on average as the team learned. Asked whether the math effort ran to hundreds of millions of dollars, he agreed, adding that the figure covers much more discovery than OpenAI has published. He confirmed OpenAI holds more than the roughly 700 proofs it has released, and said he bets the company could redo Navier Stokes at one hundredth of the cost a year after it was done.
Collison noted that mathematicians were up in arms, and Altman agreed they were not happy. He said he feels "unequivocal" that humanity deserves fundamental knowledge and that OpenAI should not sit on major discoveries, while saying the company owes it to people in those fields to find a way for them to keep a fulfilling role. He described a weekend when staff threw their hardest research problems at a new model, and said watching it solve them brought him both excitement and sadness.
Polymarket's post framed the remark as an admission amid backlash. The full transcript shows Altman presenting the week as a deliberate experiment rather than a mistake.
What to expect next
Watch whether OpenAI changes how it releases math results, for example with fuller papers, outside review or formal proofs checked in Lean before publication. Mathematicians and advisory groups may press for clearer credit and verification rules. Altman suggested more unpublished results exist, so further releases are likely. Also watch the cost claims: if similar proofs become far cheaper, other labs may run their own large agent math efforts.