OpenAI posts 700+ AI math preprints; mathematicians object

OpenAI posted more than 700 AI-generated math preprints to a GitHub repository on October 6, drawn from roughly 4,000 problems its unreleased internal model attempted. Nature reported the release covers 372 sets of results across geometry, algebra and computer science, and three manuscripts were pulled a day later after a sign error invalidated one argument and the two dependents built on it. None of the papers solved any of the five remaining Millennium Prize Problems.

The company told Nature its goal was to "evaluate and develop our models against difficult frontier problems in all sciences." The response from mathematicians was less uniform.

Alvaro Lozano-Robledo, a number theorist at the University of Connecticut, called the simultaneous release "staggering" and asked, "Why is OpenAI trying to solve hundreds of math problems to be released all at once?" Ursula Martin at the University of Oxford described the approach as "tossing the community a messy first draft and expecting us to clean it up for them." Michael Harris, a number theorist at Columbia University, told Nature he "can't bring myself to spend my time studying the outcome of this process, which I consider illegitimate."

Not every mathematician recoiled. Levent Alpöge, a mathematician at rival lab Anthropic, called the moment "the most significant moment in mathematical history," while allowing there would be "sad stories" for researchers whose projects got preempted. Three of the people we follow in our Who's Who directory circulated the Nature piece the day it ran.

What the dump means for publication norms is the open question underneath the volume. Peer review in mathematics is slow by design, and a batch of this size lands on the field at once rather than being drip-fed through journals. For now the venue of record is a GitHub repo, with three retractions already on the ledger.