OpenAI drops 722 AI math preprints; three pulled over sign error

OpenAI uploaded 722 AI-generated mathematics manuscripts to a GitHub repository on October 6, covering claimed progress on 372 open problems in geometry, algebra and computer science, Nature reported. A day later the company withdrew three of them after a sign error invalidated the argument in one paper and the constructions in two dependent manuscripts.

The scale is what rattled the field. Alvaro Lozano-Robledo, a number theorist at the University of Connecticut in Storrs, called the simultaneous posting "staggering" and asked "Why is OpenAI trying to solve hundreds of math problems to be released all at once?" Ursula Martin of the University of Oxford said the approach amounted to "tossing the community a messy first draft and expecting us to clean it up." Michael Harris of Columbia University said he would not read the preprints at all.

Not every reaction was hostile. Levent Alpöge, a mathematician at rival firm Anthropic in San Francisco, wrote on X that there would be "sad stories" for researchers whose projects were pre-empted, while describing the moment as significant for mathematics. A statement from the Association for Human Mathematics was blunter: "Releasing over 700 files at once is not a demonstration of scholarship, but a demonstration of power." The group urged colleagues to discontinue their work with OpenAI.

The release follows a September OpenAI preprint on the Navier-Stokes equations, one of the original seven Millennium Prize Problems. Beyond the three withdrawals, fourteen manuscripts in the batch have already been revised.