OpenAI announces three major scientific breakthroughs with external expert oversight
OpenAI released claims of major discoveries in mathematics, computer science, and physics, adopting a new model of expert review after internal criticism of its previous release practices.
What happened
OpenAI announced three purported scientific breakthroughs: a proof of the local smoothing conjecture for the Euclidean wave equation (resolving related Kakeya and Stein restriction conjectures in all dimensions), a proof that the Exponential Time Hypothesis is false via a deterministic subexponential-time algorithm for 3-SAT, and a mathematically rigorous existence proof of certain interacting quantum field theories including QCD and a unified completion of the Standard Model. For the computer science result, OpenAI engaged external experts Russell Impagliazzo, Ramamohan Paturi, and Ryan Williams—including two of the researchers who originally proposed ETH in 1999—to examine and explain the proof before public release. For the physics work, David Gross leads a verification group including Michael Douglas, Arthur Jaffe, Igor Klebanov, and Nikita Nekrasov. The mathematics result was released directly as AI output, with external advisers recommending this approach for researcher scrutiny. The releases include GitHub repositories, Lean proofs, and an open letter requesting community input on verification norms and broader questions about credit attribution and early access protocols for AI-generated results.
Context
This represents a shift in OpenAI's release strategy following internal and external criticism of its October mass release of hundreds of mathematical manuscripts. An AGMAI panel of mathematics advisers explicitly recommended against OpenAI's practice of granting certain people early access to results, citing unequal access and borrowed reputation concerns, and instead advocated for release on scholarly repositories with formalized proofs and transparent methodology. The computer science claim is particularly significant given that ETH has been widely believed true since 1999 and is foundational to computational complexity theory. The mathematics result directly relates to recent work for which NYU's Hong Wang received a Fields Medal just months ago for resolving the Kakeya conjecture in three dimensions, making OpenAI's claimed extension to all dimensions a notable claimed advancement. The framing of external experts as "shepherds" and the creation of verification groups signal an attempt to address credibility concerns by involving established researchers with standing in their fields.