OpenAI Is Reportedly Preparing a GitHub Math Release as Researchers Demand Papers

Mathematicians are challenging how OpenAI intends to share its next batch of AI-generated results, arguing that announcements without explanatory papers make discoveries harder to verify and credit. WIRED reported on October 6 that the company was preparing a GitHub release. OpenAI says it is working on responsible disclosure, but has not set a release time.

A publication plan, not a confirmed release

People familiar with the plans told WIRED that OpenAI intended to put hundreds of mathematical results on GitHub on Tuesday, October 6. Spokesperson Lindsay McCallum did not confirm that schedule. She said the company was drawing on advice and public recommendations from its mathematics advisory group.

The scale rests on a company claim that predates this publication dispute. In its September 21 announcement, OpenAI said an internal model had resolved more than 100 long-standing open problems across most areas of mathematics, including the Navier–Stokes Millennium Prize problem. McCallum repeated that claim to WIRED.

Researchers asked for work they could build on

In August, OpenAI gathered around 40 mathematicians to discuss AI’s potential impact on the field and how to publish findings, attendees told WIRED. Northwestern University mathematician Bryna Kra recalled a mixture of excitement and dread, but viewed the meeting as a promising first step.

Kra said attendees urged OpenAI to publish papers explaining the results, rather than disclose them through blogs or tweets. The purpose was to let mathematicians absorb, understand and use the work. She now believes that advice was ignored.

The criticism is not simply resistance to AI doing mathematics. Kra told WIRED she welcomed a powerful new tool, but wanted better disclosures so she could trust and build on its results. Other mathematicians interviewed by WIRED also objected to announcements that leave prior contributions inadequately recognized.

Math by tweet and math by press release to me is not the way to nurture the ecosystem that created the fertile ground that they have trained on

Bryna Kra, Northwestern University mathematician, speaking to WIRED

Credit disputes shadow the next disclosure

A September dispute helps explain that distrust. NYU mathematician Tristan Buckmaster accused OpenAI of front-running unpublished work he developed with Anthropic employee Levent Alpöge. Buckmaster also alleged that OpenAI researcher Sébastien Bubeck suggested excluding Alpöge from a paper. Bubeck pointed WIRED to his previous public denial that he had requested that exclusion.

OpenAI’s September advisory-group announcement acknowledged mathematicians’ concerns about using open problems as AI benchmarks. The group is meant to advise on the significance, review and communication of emerging results, including academic and professional standards.

OpenAI says the group operates independently: members are unpaid, may offer unsolicited advice and can make their advice public. Its mandate has a boundary, however. It will not advise the company on how to pace its internal progress in mathematics.