OpenAI has put 722 manuscripts on GitHub. Together they claim solutions to, or progress on, 372 major math problems. Each paper still has to survive review by mathematicians.
The company announced the release. The work comes from an unreleased ChatGPT pioneer model, and OpenAI’s new advisory group confirmed the breakthroughs. Expect the drop to widen the Navier-Stokes AI controversy that’s already roiling the mathematical community.
A single prompt, a single agent
The headline claims are big. OpenAI says it has a solution to the four-dimensional Kakeya conjecture, improvements to key computer algorithms and progress toward the hyper-challenging Riemann hypothesis.
The method is just as striking. OpenAI told Scientific American that nearly every paper came from a single prompt given to a single AI agent. The company said the “average result” took around three hours of ChatGPT Pro use.
That one-prompt, one-agent claim is exactly what skeptics are pushing back on.
Mathematicians aren’t taking OpenAI’s word for it
Nobody will know what these results are worth until mathematicians assess them. After the Navier-Stokes imbroglio, a lot of scientists are still doubtful.
“Until and unless they release the model and people can replicate their results, I think you should treat any claims about one-shotting problems with a single agent as unverified,” MIT mathematician Andrew Sutherland told Scientific American.
That’s a fair bar. The model behind these papers isn’t public, so for now nobody outside OpenAI can rerun the work.
OpenAI followed some of its advisers’ rules and skipped others
OpenAI said it released the papers under guidelines recommended by its independent Advisory Group on Mathematics and Artificial Intelligence (AGMAI). Those guidelines call for prompt release of results through traditional academic channels, along with details like the name of the model used, the prompts and the compute costs.
The release includes the model’s reasoning, compute estimates and information on how many problems were attempted. But OpenAI ignored some of the board’s suggestions. It didn’t publish specific compute times for individual problems, and it didn’t disclose which prompts it used.
Those gaps matter because the prompts and per-problem compute are the details outside researchers would need to check the single-prompt claim.
“For this release, we’re publishing the results in a GitHub repository, with protocols for paper revisions and citations,” the company said. “We’re continuing to explore other community-hosted alternatives for this release which meet the committee’s guidelines. For future releases, we are committed to further improving the quality of the papers via the citations, mathematical exposition, and presentation of the results for better understanding.”
The drop was no surprise
None of this came out of nowhere. Last month OpenAI said its models had resolved “more than 100 long-standing open problems across most areas of mathematics.” This release is the paperwork behind that claim, and then some.
If you want to judge it yourself, skip the press release and go to the GitHub repository. Start with the Kakeya paper, then watch how working mathematicians respond to it over the coming weeks. Until someone outside OpenAI reproduces even one of these results, Sutherland’s word for them is the right one: unverified.
