OpenAI posts 722 AI math preprints, withdraws three next day

OpenAI uploaded 722 AI-generated mathematics preprints to a GitHub repository on October 6, each claiming progress on an open math problem, covering 372 results in geometry, algebra and computer science. The model that wrote them has not been released. Nature's write-up notes that only 162 of the 722 manuscripts had their main results formalized in Lean; the rest went up without independent verification.

A day later OpenAI pulled three. A sign error invalidated the argument in one paper and the constructions in two others that depended on it. Fourteen more were revised.

The reaction from working mathematicians was not warm. "I can't bring myself to spend my time studying [these preprints], so I will not be reading these papers," said Michael Harris, a number theorist at Columbia University. Ursula Martin, a mathematician at the University of Oxford, told Nature that OpenAI was "tossing the community a messy first draft and expecting us to clean it up for them." Alvaro Lozano-Robledo, a number theorist at the University of Connecticut, asked: "Why is OpenAI trying to solve hundreds of math problems to be released all at once?"

A group letter circulating on social media put it more flatly: "Releasing over 700 files at once is not a demonstration of scholarship, but a demonstration of power."

Not every mathematician was hostile. Levent Alpöge, who works at rival lab Anthropic, posted on X that whatever one thinks of the drop, "it's obviously the most significant moment in mathematical history." OpenAI's stated aim is to "provide tools for mathematicians and scientists." Two researchers in our Who's Who tracker shared the Nature piece within hours of publication, which gives a sense of how fast it moved through the mathematical-reasoning crowd we follow.

The batch did not include solutions to the five remaining Millennium Prize Problems. OpenAI's prior math showpiece, a claimed proof touching the Navier-Stokes equation, came in September.