OpenAI Announces Three Scientific Discoveries and a New Model for Working with Academics

Less than a month after releasing a large batch of research preprints it said contained solutions to long-standing mathematical problems, OpenAI has returned to announce three results it describes as "major" across three distinct fields: mathematics, theoretical computer science, and theoretical physics. But the most striking element this time is not the substance of the results alone, but rather the way the company has chosen to manage the announcement and verification process, by involving teams of prominent researchers in reviewing the outputs and explaining them to the scientific community.

Three Results in Three Fields

The first result concerns combinatorics and harmonic analysis, where OpenAI says its model arrived at a proof of the Local Smoothing conjecture for the Euclidean wave equation in all spatial dimensions starting from dimension two. According to the company, this proof is supposed to entail other related results, among them the Kakeya conjecture and Stein's estimates for spherical restriction within the expected range. It is worth noting that researcher Hong Wang of New York University received the Fields Medal a few months ago for solving the Kakeya conjecture in three dimensions, which illustrates the magnitude and standing of the problem.

The second result is in computer science, and is the most controversial: a paper claiming that the Exponential Time Hypothesis is false. The paper presents a deterministic algorithm that exactly counts the satisfying assignments of fixed-width CNF formulas in sub-exponential time and polynomial space, and the 3-SAT case of it refutes the hypothesis. This hypothesis has been widely accepted among specialists since it was proposed in 1999, and is used as a foundation for many computational complexity results, meaning that refuting it would have far-reaching consequences.

The third result is in theoretical physics, and has not yet been published in full. OpenAI describes it as a "mathematically rigorous existence proof" for certain interacting quantum field theories containing gauge fields and fermions in four dimensions, including quantum chromodynamics and a unified completion of the Standard Model of particle physics.

"Shepherds": A New Review Model

What is new in this round is that OpenAI did not publish the outputs on its own, but instead entrusted the follow-up verification to groups of outside experts it calls "shepherds":

• Computer science: an independent working group led by Russell Impagliazzo and Ramamohan Paturi of the University of California San Diego — who originally proposed the hypothesis — with the participation of Ryan Williams of MIT, to examine the proof and explain it.
• Physics: a team led by David Gross and including Michael Douglas, Arthur Jaffe, Igor Klebanov and Nikita Nekrasov.
• Mathematics: the model's outputs were published directly, as in the previous announcement, based on advice holding that researchers should arrive at their own interpretations of the result.

At the request of the computer science group, the materials were made publicly available immediately, including the output summary and the proof formalized in Lean on GitHub, so that others could examine them for themselves.

Objections to "Early Access"

Even so, the company acknowledges that this approach may carry its own problems. The mathematics advisory committee that OpenAI itself formed recommended that AI companies not adopt the practice of designating specific individuals to receive outputs first, citing three concerns: unequal access, the "borrowing" of scientific reputation to bolster the credibility of results, and biased selectivity in choosing reviewers.

Instead, the committee proposed that AI labs publish their outputs on open academic repositories, with proofs formalized in Lean, along with clarification of how the problems were selected, and funding for researchers working to understand and verify the results.

Open Questions for the Scientific Community

The announcement was accompanied by an open letter requesting input on how to scrutinize these claims and establish norms for dealing with AI-generated "breakthroughs." Among its questions:

• What rules should govern early access to AI-generated results, disclosure of conflicts of interest, and the use of unpublished materials in subsequent research?
• How should scientific credit be distributed among those who discover and those who verify, correct, simplify, explain and extend?
• How do we support students and researchers as the tools change, while preserving the understanding and critical judgment needed to hold results to account?

The Practical Value of What Is Happening

The significance of this announcement is not limited to the problems themselves. We are witnessing a genuine experiment in a new mechanism for producing knowledge: a model generates a proof, humans verify it, and a scientific community needs rules that have not yet been written. Until the reviews are completed, these results remain claims under examination rather than settled facts — a fundamental distinction that must be maintained no matter how large the announcing party may be.

✦ بقلم فريق دروب أيديا