Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseAPI / Product · ActionableThe Verge→ original

OpenAI Announced Three Scientific Breakthroughs, Kept Physics Under Wraps, and Picked the Reviewers Itself

OpenAI announced three scientific results, including a solution to the Kakeya needle problem about rotating a needle in all directions. The math proof was released openly, the computer science result will be handed to an external group, and the physics result was not published at all, with the company choosing the reviewers itself. Scientists call the results "huge" but disagree with the closed circle of reviewers. If you publish conclusions from AI agents, release the materials to everyone at once.

AI-processed from The Verge; edited by Hamidun News
OpenAI Announced Three Scientific Breakthroughs, Kept Physics Under Wraps, and Picked the Reviewers Itself
Source: The Verge. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceThe Verge ↗
Editor-in-Chief
The exponential time hypothesisOpenAI says it has disproved it
Key metric #2Independent mathematicians from AGMAI are asking for the physics proofs to be opened
Key metric #3Physicist Andreas Karch called the hand-picked choice of reviewers a "catastrophe"

Imagine your team ships a major release and shows the code only to a few well-known reviewers you picked yourself. Other developers wait and cannot understand why they are not allowed to look. A similar dispute has erupted around OpenAI. On Monday, the company announced three results at once in mathematics, theoretical computer science, and physics, and said it is changing its approach to publication. The scientists surveyed by The Verge consider the results themselves serious. But they are tired of OpenAI's style: in early October, the company dumped hundreds of preprints on mathematicians, and they are still working through them.

What OpenAI Actually Claimed

The first result is mathematical. The company says it has solved the Kakeya conjecture in all dimensions. Put simply, the problem asks how little space is needed to turn a needle or a pencil in every direction. A proof for the three-dimensional case was among the achievements for which mathematician Hong Wang of New York University received this year's Fields Medal. Brendan Hassett, a professor at Brown University, said the claim "looks well-founded." But he added a caveat: he has not yet examined the formalization in Lean, a proof written as code that a computer checks step by step.

The second result is in computer science. OpenAI says it has disproved the exponential time hypothesis. This is a widely held assumption: some problems cannot be solved significantly faster than by trying every possible option. It is closely tied to the P versus NP problem, which asks whether a problem whose answer is easy to verify is also always easy to solve.

The third result is in physics. It concerns quantum chromodynamics, which describes the strong nuclear force, and its connection to Yang–Mills theory. The company clarifies that this work does not touch the unsolved "mass gap" problem on the list of Millennium Prize Problems.

Why Mathematicians and Physicists Are Unhappy

For each result, OpenAI chose its own publication method, following the advice of "shepherds", invited experts. The AI-produced mathematical argument was released in full so that researchers could organize their own review. For computer science, the company will assemble an external group to explain the result. The physics result was not published at all: OpenAI promised to hand the decision on how to release it over to outside experts.

This is precisely what irritates people. Konstantin Kogler of the Institute for Advanced Study is not opposed to there being three results now rather than hundreds. But, in his words, there is no point in showing the work first only to a few eminent scientists. Physicist Andreas Karch of the University of Texas at Austin called the hand-picked review group a "catastrophe": large labs already have enough power, and they should not be appointing winners and losers.

A day before the announcement, the Advisory Group on Mathematics and AI (AGMAI), whose advice OpenAI had promised to heed, recommended that companies not use this practice. In its view, selected recipients are unnecessary, their reputations are simply being lent to lend credibility, and the practice risks unequal access, biased selection, and delays. The group advises posting results to scientific repositories along with formalized proofs and an explanation of how the problems were selected, and supporting those who carry out the review. After the announcement, AGMAI demanded that the physics materials be opened without delay. The argument is simple: according to OpenAI itself, the Lean formalization is already complete, so there is no scientific reason to hold back.

What This Means for Anyone Releasing AI Results

The first lesson: machine verification such as Lean helps people trust AI results, but it does not replace open access. Hassett notes that when groups around the world work through results, it is easier for everyone else to start using the new tools. Stanford mathematician Jared Duker Lichtman sees the other side too: it is impossible to please everyone, so the goal is to find the least offensive option.

The second lesson runs deeper. Xiaoyu He of the Georgia Institute of Technology argues that the problem is not the announcement format. Fundamental problems are being solved too quickly and without human involvement, and the papers are hard to read and check. The cost of verification falls on the human community. If your agents produce results that need checking, build open artifacts, clear descriptions, and resources for external review into the process.

There is also an honest caveat. None of the three results has been independently confirmed yet. The physics materials are closed, and the Lean proof for mathematics still has to be studied. Experts call the results "huge" and "serious breakthroughs," but the word "breakthrough" should be read as the company's claim, one that will be confirmed or refuted in the coming months.

Story

Key Questions Answered

What OpenAI Actually Claimed?

The first result is mathematical. The company says it has solved the Kakeya conjecture in all dimensions. Put simply, the problem asks how little space is needed to turn a needle or a pencil in every direction. A proof for the three-dimensional case was among the achievements for which mathematician Hong Wang of New York University received this year's Fields Medal.

Why Mathematicians and Physicists Are Unhappy?

For each result, OpenAI chose its own publication method, following the advice of "shepherds", invited experts. The AI-produced mathematical argument was released in full so that researchers could organize their own review. For computer science, the company will assemble an external group to explain the result. The physics result was not published at all: OpenAI promised to hand the decision on how to release it over to outside experts. This is precisely what irritates people.

What This Means for Anyone Releasing AI Results?

The first lesson: machine verification such as Lean helps people trust AI results, but it does not replace open access. Hassett notes that when groups around the world work through results, it is easier for everyone else to start using the new tools. Stanford mathematician Jared Duker Lichtman sees the other side too: it is impossible to please everyone, so the goal is to find the least offensive option. The second lesson runs deeper.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…