OpenAI to Dump 100+ Math Solutions, Angering Researchers
Original: OpenAI Is Pissing Off a Bunch of Mathematicians—Again
Why This Matters
AI firms bypassing peer review could upend how mathematics is validated and credited.
OpenAI plans to release 100+ solutions to unsolved math problems on GitHub, defying mathematicians' explicit requests for peer-reviewed papers. Academics say AI firms are treating math as a PR arena ahead of their IPOs.
OpenAI gathered roughly 40 mathematicians in August to discuss what happens when AI surpasses human capability in the field. The company indicated its models had solved hundreds of long-standing problems, and attendees say they were told solutions wouldn't be dumped all at once—an assurance OpenAI spokesperson Lindsay McCallum says the company is "not aware of." Northwestern's Bryna Kra says the group specifically requested proper papers, not blog posts or tweets. "Apparently, that input was ignored," she told WIRED.
OpenAI now plans to release more than 100 results on GitHub, including work tied to the Navier-Stokes Millennium Prize problem. McCallum confirmed the model has "resolved more than 100 long-standing open problems across most areas of mathematics" but said no release date is set.
The deeper frustration is about scientific norms. In September, OpenAI deployed thousands of agents to solve a Millennium Prize problem after hearing "rumors" that competitors were close. NYU mathematician Tristan Buckmaster accused the company of front-running work he had done in a personal collaboration with Anthropic employee Levent Alpöge—work they hadn't yet published. During credit negotiations, Buckmaster claims an OpenAI researcher suggested Alpöge should be excluded from any resulting paper. Mathematicians say both OpenAI and Anthropic are using the field to score points ahead of blockbuster IPOs.