OpenAI to Dump Hundreds of Solved Math Problems on GitHub

OpenAI is preparing to upload hundreds of solutions to unsolved mathematical problems to GitHub. The upcoming release follows a series of AI-driven breakthroughs, including the training of a new internal model that began on August 28. According to OpenAI spokesperson Lindsay McCallum, this model has successfully resolved the Navier-Stokes Millennium Prize problem alongside more than 100 other long-standing open mathematical problems.

The impending dump has frustrated the academic community, which accuses the AI giant of bypassing traditional scientific verification. In August, OpenAI hosted about 40 mathematicians to discuss the implications of AI outpacing human capabilities. Attendees like Northwestern University mathematician Bryna Kra urged the company to publish detailed papers rather than releasing results via blog posts or tweets. Kra noted that this advice was seemingly ignored, arguing that "math by press release" fails to nurture the academic ecosystem.

Tensions escalated in September when OpenAI deployed thousands of agents to tackle a million-dollar Millennium Prize problem. New York University professor Tristan Buckmaster accused OpenAI of front-running research he was conducting with Anthropic employee Levent Alpöge. Buckmaster claimed OpenAI researcher Sébastien Bubeck suggested excluding Alpöge from authorship and warned Buckmaster against speaking to the press. Bubeck has denied these claims, but the incident has intensified fears of "mobster behavior" from dominant AI firms, according to NYU visiting professor Nestor Guillen.

In response to the influx of machine-generated proofs, researchers have established new verification tools, including the Hexagon repository and the Palomar registry. Over 4,000 mathematicians have also signed the Leiden declaration, demanding AI companies adhere to academic standards. While some OpenAI employees reportedly believe AI has rendered the field of mathematics dead, academic researchers argue the technology should instead be used to expand human capabilities, provided the findings are released responsibly.