OpenAI dumps 700+ AI math preprints; mathematicians object

On October 6, OpenAI posted more than 700 AI-generated mathematics preprints to a public GitHub repository, written by an unreleased internal model that had been pointed at roughly 4,000 problems. The company says the release covers 372 sets of results across geometry, algebra and computer science, and told Nature its goal was to 'evaluate and develop our models against difficult frontier problems in all sciences.'

Working mathematicians were not pleased. Alvaro Lozano-Robledo, a number theorist at the University of Connecticut in Storrs, 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, a mathematician 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 in New York City, told Nature he 'can't bring myself to spend my time studying' the preprints.

Not everyone recoiled. At rival lab Anthropic, mathematician Levent Alpöge called the release 'obviously the most significant moment in mathematical history,' while also noting 'sad stories' from researchers whose in-progress projects had been pre-empted.

A day after posting, OpenAI withdrew three manuscripts. A sign error in a stabilization-trace cancellation argument inside 'Algebraicity of Weil classes on split abelian eightfolds' invalidated that paper and the constructions in two dependents. None of the dump solved any of the five remaining Millennium Prize Problems.

A group of mathematicians put the structural objection flatly: 'Releasing over 700 files at once is not a demonstration of scholarship, but a demonstration of power.'