OpenAI published 719 mathematical manuscripts across 372 problem families to a public GitHub repository on Oct. 6, generated by an internal model and covering algebra, geometry and theoretical computer science, according to the repository.
The model worked through roughly 4,000 problems, with each surviving result averaging about three hours of ChatGPT Pro compute, and about 42% of the top-line results carry a proof formally verified in the Lean proof assistant, OpenAI said. The repository's own documentation warns the results are "at different stages of verification" and that "some of the unformalized results could have issues," with the company committing to post corrections.
The Association for Human Mathematics, in a statement, called the release "not a demonstration of scholarship, but a demonstration of power" and urged mathematicians to stop working with OpenAI. Fields Medalist Hugo Duminil-Copin told The Decoder, "The shock is immense. I am paralysed," and Cambridge mathematician Henry Wilton said, "If OpenAI wanted to destroy the mathematical community, this would be a great way."
Other mathematicians pushed back on the alarm. Fields Medalist Peter Scholze urged patience, saying "the goal is and always will be the human understanding of mathematics," according to The Decoder, and Terence Tao described mixed feelings, writing that some of the AI-generated proofs "introduce clever new ideas" even as he lamented the lack of human collaboration behind them.
A single, unreviewed GitHub repository has become one of the most contested events in recent mathematics, and the fight is as much about who gets credit and jobs in a field AI can now contribute results to as it is about whether the individual proofs hold up.