OpenAI dumps 722 AI math preprints to GitHub, pulls three

OpenAI uploaded 722 preprints to a GitHub repository on October 6, each claiming progress on an open math problem. The drop covers 372 sets of results in geometry, algebra and computer science, produced by an unreleased internal model, Nature reported. Roughly half were released without independent verification.

Within a day, three were pulled. A sign error invalidated an argument in one manuscript and the constructions in two dependent papers; fourteen others received revisions, Retraction Watch wrote.

'Why is OpenAI trying to solve hundreds of math problems to be released all at once?' asked Alvaro Lozano-Robledo, a number theorist at the University of Connecticut in Storrs, who called the simultaneous drop 'staggering' on X. Levent Alpöge, a mathematician at Anthropic, said on X there would be 'sad stories' for researchers whose projects might be pre-empted, while also calling it 'the most significant moment in mathematical history.'

A group letter circulating on social media put it more flatly: 'Mathematicians did not ask for this work to be done,' it reads, and 'Releasing over 700 files at once is not a demonstration of scholarship, but a demonstration of power.'

Alex Townsend of Cornell argued OpenAI should have led with the subset verified in Lean. 'Given the skepticism of AI in the mathematics community, I think that OpenAI should have announced the manuscripts that were lean verified first,' he said. Andrew Sutherland of MIT called the fast withdrawal 'the responsible thing to do' but added that 'it will take a lot more than that to earn back the trust they have lost.'

OpenAI said it 'welcome[s] scrutiny and feedback from the mathematical community' and will withdraw papers 'if no fixes can be found.' The company's stated goal, it said, is to 'provide tools for mathematicians and scientists.' None of the five remaining Clay Millennium Prize Problems appear among the preprints.