OpenAI drops 700 AI math preprints; mathematicians object

OpenAI uploaded more than 700 AI-generated mathematics preprints to GitHub on October 6, written by an unreleased internal model and claiming progress on 372 open problems in geometry, algebra and computer science, Nature reported.

The mathematics community did not take it well. Michael Harris, a number theorist at Columbia University, called the process "illegitimate" and told Nature, "I will not be reading these papers." Ursula Martin, a mathematician at the University of Oxford, said OpenAI was "tossing the community a messy first draft and expecting us to clean it up for them." One researcher labelled the moment a "scoopolypse," worried that human mathematicians would find in-progress projects pre-empted by the bulk release.

Not everyone objected. Levent Alpöge, a mathematician at Anthropic, called the release "the most significant moment in mathematical history." An OpenAI spokesperson said the company's goal was to "provide tools for mathematicians and scientists" by evaluating its models against difficult frontier problems in all sciences.

The follow-through was rougher. A day after posting, OpenAI withdrew three of the manuscripts after a sign error invalidated the argument in one paper and the constructions in two dependents, Retraction Watch reported. None of the preprints claimed a solution to any of the remaining Millennium Prize Problems. Three AI watchers on our radar passed around the Nature piece.

"Why is OpenAI trying to solve hundreds of math problems to be released all at once?" Alvaro Lozano-Robledo, a number theorist at the University of Connecticut, asked on X.