OpenAI posts 722 AI math preprints; mathematicians balk

On October 6, OpenAI uploaded 722 AI-generated math preprints to a GitHub repository, covering purported progress on 372 open problems in geometry, algebra and computer science, produced by an unreleased internal model that was pointed at roughly 4,000 problems, Nature reports.

Only 162 of the 722 manuscripts had their main results formalized in Lean. A day later OpenAI withdrew three, after a sign error invalidated one argument and the constructions in two dependent papers fell with it. Fourteen others were revised with proof repairs.

Alvaro Lozano-Robledo, a number theorist at the University of Connecticut, called the simultaneous drop 'staggering' on X and asked, 'Why is OpenAI trying to solve hundreds of math problems to be released all at once?' A group letter circulating on social media was blunter: 'Releasing over 700 files at once is not a demonstration of scholarship, but a demonstration of power.'

Ursula Martin, at the University of Oxford, likened the release to 'tossing the community a messy first draft and expecting us to clean it up for them.' Michael Harris, a number theorist at Columbia University, was flatter: 'I can't bring myself to spend my time studying [these preprints], so I will not be reading these papers.' Not every reaction was hostile. Levent Alpöge, a mathematician at Anthropic, acknowledged potential harm to researchers whose projects could be preempted, but called the release 'the most significant moment in mathematical history' for what it demonstrated about AI's usefulness to working mathematicians.

An OpenAI spokesperson told Nature the goal is to 'provide tools for mathematicians and scientists' by evaluating models against 'difficult frontier problems in all sciences.' The batch did not solve any of the five remaining Millennium Prize Problems.