OpenAI posts 700+ AI math preprints, pulls three a day later

OpenAI uploaded more than 700 AI-generated mathematical preprints to GitHub on 6 October, produced by an unreleased model, Nature reports. The simultaneous dump has drawn an angry reaction from working mathematicians.

Alvaro Lozano-Robledo, a number theorist at the University of Connecticut in Storrs, called the volume "staggering" and asked on X: "Why is OpenAI trying to solve hundreds of math problems to be released all at once?" Oxford's Ursula Martin was blunter, telling Nature that OpenAI was "tossing the community a messy first draft and expecting us to clean it up for them."

Columbia number theorist Michael Harris declined to engage at all. "I can't bring myself to spend my time studying [these preprints]," he said, adding that he would not be reading the papers.

Not every reaction was hostile. Anthropic's Levent Alpöge acknowledged "sad stories" for researchers whose projects might be preempted, yet called the release "the most significant moment in mathematical history" for mathematical technology. An OpenAI spokesperson said the company's goal was evaluating models "against difficult frontier problems in all sciences."

The release did not include solutions to the five remaining Millennium Prize Problems. A day later, Retraction Watch reported, OpenAI withdrew three manuscripts after a sign error invalidated one paper's argument and "the construction used by two dependent papers," and revised 14 others.