How far can artificial intelligence reach into problems that scientists have wrestled with for decades? OpenAI's latest announcement gives that question a whole new dimension. The company says its models have produced three theoretical breakthroughs it describes as "major" across mathematics, computer science and physics — and the announcement came less than a month after the company released hundreds of mathematical solutions to the public.

Another notable detail is how the process was managed. OpenAI said it appointed independent external experts as "shepherds" to guide the findings in computer science and physics. In mathematics, meanwhile, the model's result was made available on GitHub so that researchers could build on it in their own work.

openaiic

A core assumption of computer science shaken

According to OpenAI's claim, one of computer science theory's most entrenched assumptions — the Exponential Time Hypothesis (ETH) — no longer holds. The company's published paper describes a deterministic algorithm that counts assignments satisfying constant-width CNF formulas in sub-exponential time and polynomial space, one by one. Applying this algorithm to the 3-SAT problem, the claim goes, explicitly refutes ETH, which had been accepted as true for decades.

Verification was taken seriously. Russell Impagliazzo and Ramamohan Paturi of UC San Diego — two of the figures who brought the hypothesis to prominence in 1999 — together with Ryan Williams of MIT, led an independent working group. At the group's request, OpenAI shared the model's output and a formal proof in the Lean language publicly on GitHub.

A mathematics claim in the shadow of the Fields Medal

In the mathematics portion, OpenAI claims the work proved Sogge's local smoothing conjecture for waves in two and all higher spatial dimensions, via an auxiliary partial differential equation. According to the company, this proof also resolves the Kakeya maximal and dimension conjectures and Stein's spherical restriction estimates in all dimensions. As a reminder, the Kakeya conjecture, in geometric measure theory, concerns the smallest area or volume in which a unit-length needle can be rotated.

There is a striking backdrop here: Hong Wang of NYU received the prestigious Fields Medal from the International Mathematical Union just a few months ago for solving the Kakeya conjecture in three dimensions only. OpenAI, by contrast, claims its model solved the same problem in all dimensions.

Physics: incomplete data, but a heavy card

The physics results have not yet been fully disclosed. Based on the available information, the work presents a mathematically rigorous existence proof for interacting quantum field theories in four dimensions containing gauge fields and fermions. The claim is that this proof covers Quantum Chromodynamics and a unified completion of the Standard Model of particle physics. An expert panel consisting of Michael Douglas, Arthur Jaffe, Igor Klebanov and Nikita Nekrasov took part in the work under the leadership of David Gross.