Researchers Push Back as OpenAI Limits Access to Its Math, Physics and Computer Science Results

On Monday, OpenAI announced breakthroughs in mathematics, computer science and physics, and said it was rethinking how it communicates its findings. The company's new approach came after a rocky few months with mathematicians, who are still digesting the hundreds of preprints the company dropped on them in early October. Researchers who spoke to The Verge say the course correction has not won them over.

What results did OpenAI announce in math, physics and computer science?

The three results tackle problems that sit near the center of their fields. Some build on OpenAI's earlier work, while others break new ground. The company says it consulted expert “shepherds” before deciding how to release each one, and each result has a different release plan.

Is OpenAI's Kakeya conjecture claim well supported?

The mathematics result is the most eye-catching. It claims to resolve the Kakeya conjecture in all dimensions. The problem asks, roughly, how little space is needed to rotate a needle or pencil in every direction. NYU mathematician Hong Wang received a Fields Medal earlier this year, partly for proving the three-dimensional version. Brendan Hassett, a mathematics professor at Brown University, told The Verge that OpenAI’s claim “looks well supported.” He said he had not yet checked the Lean formalization, a tool for verifying results by computer.

OpenAI released the mathematics output itself and left researchers free to explain it on their own. The physics result is different. It has not been released at all. OpenAI said it will “defer” to an external group of experts on how to share those materials.

The physics work examines the Standard Model, the most successful and complete theoretical framework in physics. It focuses on quantum chromodynamics, which describes the strong nuclear force, and its link to Yang-Mills theory, a cornerstone of modern particle physics. Yang-Mills theory also sits at the heart of one of the seven Millennium Prize problems. OpenAI’s work does not address the unresolved “mass gap” in that problem.

The final result claims to disprove the exponential time hypothesis, a widely held assumption in theoretical computer science that certain problems cannot be solved much faster than by brute force. The result is closely tied to P versus NP, another Millennium Prize problem and one of the field’s biggest open questions.

Judging the work is hard. Researchers say the varying levels of disclosure make a full evaluation difficult, and even with complete information, checking the claims would take time. Still, the experts who spoke to The Verge said there are good reasons to take the results seriously. They called them “huge,” “major breakthroughs” and “significant.”

The excitement came with plenty of exasperation. Researchers described months of what they called an exhausting stream of AI-generated findings. Many want the wider community to start the slow work of checking OpenAI’s claims, and they question why the company is adding barriers by choosing who sees the work first.

Why are researchers pushing back on OpenAI's release plan?

Most of the experts welcomed OpenAI’s decision to consult outside researchers. They argued, however, that if a company announces results publicly, the work behind them should also be available for anyone to examine. Constantin Kogler, a researcher at the Institute for Advanced Study, said he was fine with the release of three problems instead of hundreds. He objected to restricting access to a group of OpenAI’s choosing. “It doesn’t make any sense that first only a few distinguished scientists can look at them,” he said.

Andreas Karch, a physics professor at the University of Texas at Austin, was blunter. “The fact that they handpick the group is a disaster,” he told The Verge. “The large labs have enough power as it is. They should not be picking winners and losers.” Karch suggested a “competitive process” for choosing reviewers, though he acknowledged it would probably produce a similar group, which he called “first rate.” He also asked why OpenAI “can’t release the proof now” alongside naming its experts. “This would democratize access while still showing that they care about making the results meaningful and understandable,” he said.

The criticism echoes a warning from the Advisory Group on Mathematics and Artificial Intelligence, or AGMAI. OpenAI had pledged to consult the group on how to share its findings. A day before the announcement, the group said it recommended that AI companies avoid this practice. Giving selected people the output first, it argued, is unnecessary, lends them the reputation of the chosen researchers, and risks unequal access, biased selection and delays.

AGMAI recommended that AI labs post their output to scholarly repositories, include formal proofs and explain how problems were chosen. It also called for funding and other support for researchers trying to understand the results. After the latest announcement, the group urged OpenAI to make public “without delay” the natural-language and Lean outputs behind its physics claims. It said OpenAI had delayed release until the Lean formalization was complete, and by the company’s own account that work is now done.

Hassett said open release could also improve the science. “Having many groups around the world wrestle with these results supports rapid adoption of new tools by those that would like to use them,” he said.

Xiaoyu He, a mathematics professor at the Georgia Institute of Technology, said the problem runs deeper than the announcement format. “The results are very impressive,” he said, “but I think they are making a show of trying to be responsible and thoughtful about how they release results while not being able to change the main issues: that fundamental results are being solved so quickly without human contribution, and that the papers are difficult to read and verify.”

OpenAI says it is experimenting with different approaches and acknowledges that each one has tradeoffs. Stanford mathematician Jared Duker Lichtman summed up the dilemma. “My sense is that for any approach, there is going to be someone who is upset,” he said. “So it’s just a question of finding the least offensive option.”

This article was written with the assistance of AI.
News Factory APP - agentic news to boost your SEO & AEO.