OpenAI says its models have produced three major results in mathematics, computer science, and physics. The announcement also marks a change in how the company is handling some of its AI research: outside specialists will help guide the review of two results, while the mathematics output is being shared for researchers to assess themselves.
Three claims, three research areas
The results cover problems with very different kinds of proof. In mathematics, OpenAI says a model produced a proof of the local smoothing conjecture for waves in every dimension. The company says this also resolves the Kakeya conjecture and Stein’s restriction conjecture in all dimensions.
OpenAI has published the mathematical output directly, as it did in an October 6 release. Its description says the proof uses an auxiliary partial differential equation to establish local smoothing for the Euclidean wave equation in every spatial dimension at least two. It also says the work yields the Kakeya maximal and dimension conjectures, along with Stein’s spherical restriction estimates throughout the interior of the conjectured range.
The computer science claim concerns the Exponential Time Hypothesis, or ETH, which the article describes as widely believed to be true. OpenAI says its paper presents a deterministic algorithm that exactly counts satisfying assignments of fixed-width CNF formulas in subexponential time and polynomial space. The 3-SAT case, it says, refutes ETH.
The physics result is described as a mathematically rigorous existence proof for certain interacting quantum field theories in four dimensions. Those theories contain gauge fields and fermions, and include Quantum Chromodynamics (QCD) and a unified completion of the Standard Model of particle physics. OpenAI has not yet released the full physics information.
Outside experts are being asked to guide review
For the computer science result, an independent working group will examine and explain the proof. It is led by Russell Impagliazzo, Ramamohan Paturi, and Ryan Williams. At the group’s request, OpenAI says it is making the materials public immediately so that others can inspect them too.
The computer science review includes experts Russell Impagliazzo and Ramamohan Paturi, who were part of the group that originally proposed ETH in 1999. The writeup and Lean proof are being released on GitHub. Lean is also mentioned in recommendations from math advisers as a way to formalize proofs for scholarly review.
For the physics work, OpenAI says David Gross will lead a group including Michael Douglas, Arthur Jaffe, Igor Klebanov, and Nikita Nekrasov. The company describes these outside experts as “shepherds” guiding the work. Their role signals a process for examining the result, but the article does not report that the claims have been independently confirmed.
A different route for the mathematics result
The mathematics output is being handled differently. OpenAI says its advisers recommended releasing the model’s output directly so researchers can work through their own explanations. That leaves the task of understanding and assessing the proof open to the wider research community.
The stakes are visible in the surrounding context: just a few months ago, NYU professor Hong Wang was awarded a Fields Medal by the International Mathematical Union for resolving the Kakeya conjecture in three dimensions. OpenAI’s claim goes further in scope, saying its model’s proof addresses the conjecture in all dimensions. That comparison helps explain why researchers may want to examine the details closely; it does not establish whether the new proof holds up.
Release practices remain contested
Using selected experts to receive or review results first can help organize scrutiny, but it can also raise concerns about who gets access and whose reputation is attached to a claim. The article notes that an AGMAI panel of math advisers had just recommended that AI companies not designate certain people to receive output first, citing unequal access, borrowed reputation, and biased selection.
The panel instead recommended that AI labs release results on scholarly repositories, provide Lean-formalized proofs and information about how problems were chosen, and fund people working to understand the results. Those proposals point to the larger issue raised by OpenAI’s announcement: producing a striking AI result is only one step. Researchers also need access to the material and a way to scrutinize the reasoning behind it.
For now, the three claims are at different stages of public examination. The mathematical output is available directly, computer scientists are organizing review of the ETH result, and the full physics material has not been released. How the proofs withstand scrutiny will determine what the announcements mean for mathematics, computer science, and physics.