OpenAI says its AI models produced new results in mathematics, computer science and physics

OpenAI announced on Monday that its AI models had produced three new results, in mathematics, computer science and physics. The reports reviewed by Vocemundi do not describe independent verification of any of them.

In mathematics, OpenAI says its models proved the local smoothing conjecture in harmonic analysis in every spatial dimension. In computer science, OpenAI presents a result as disproving the Exponential Time Hypothesis. In physics, it presents a mathematically rigorous existence proof for quantum field theories, including quantum chromodynamics (QCD).

The AI-generated output for the mathematics and computer science results was made public with the announcement. The physics results were not released in full. OpenAI said outside experts, which it calls "shepherds", are guiding how some results are released and examined.

Reported details

According to Al Jazeera, researchers had previously proved the local smoothing conjecture only in three dimensions. According to The Verge, OpenAI's write-up says the proof covers every spatial dimension of at least two.

According to The Verge, a working group of Russell Impagliazzo and Ramamohan Paturi of UC San Diego and Ryan Williams of MIT will examine the computer science proof. The Verge says the group released the write-up and a Lean proof on GitHub, with an open letter seeking input on norms for AI-generated results.

The Verge quoted OpenAI as saying: "David Gross will lead a group including Michael Douglas, Arthur Jaffe, Igor Klebanov, and Nikita Nekrasov that will shepherd this work." According to Al Jazeera, OpenAI said the physics group deferred release to write up its own interpretation.

What is disputed

The two outlets differ on the mathematics result. Al Jazeera, citing OpenAI, says OpenAI consulted three shepherd groups. The Verge names shepherd groups only for computer science and physics. It says advisers recommended releasing the mathematics output directly.

What each side says

OpenAI says expert collaborations help place the results in context. "All of these approaches involve tradeoffs," OpenAI said, according to Al Jazeera. It added that choosing initial shepherds "creates asymmetry of information" and said it welcomes feedback.

According to The Verge, a panel of mathematics advisers assembled by OpenAI, called AGMAI, recommended a day earlier that AI companies not give designated people first access to output. The Verge says the panel cited unequal access, borrowed reputation and biased selection, and urged release on scholarly repositories instead.

Alex Iosevich, a mathematics professor at the University of Rochester, told Al Jazeera: "The conjecture was formulated by my PhD adviser, Chris Sogge, so it is near and dear to my heart, and the news was a shock at first." Al Jazeera says he called it a historic development. The reports do not indicate whether he examined the proof himself.

According to The Verge, OpenAI released hundreds of manuscripts on 6 October that it said solved long-standing mathematics questions.

Analysis

The announcement rests on OpenAI as the sole origin of the scientific claims. Expert review is still at an early stage. If the named groups and wider research communities confirm the proofs, the results could be significant for their fields. If they find gaps, attention may shift to how such claims are announced. The process itself is also under debate. OpenAI's own advisory panel questioned early-access shepherding, and the computer science group has asked for community input on norms. How OpenAI responds, and whether the physics work is released in full, may shape how AI-generated research is reviewed.

Produced by the Vocemundi newsroom with AI assistance.