Our article2 min read
OpenAI says its AI disproved a famous computer science idea. Experts are checking
OpenAI claims its models disproved the Exponential Time Hypothesis and proved two other major results. None is peer reviewed yet. Here is what is claimed and how it is being checked.
In this article
Key takeaways
- OpenAI says an internal model disproved the Exponential Time Hypothesis, a computer science conjecture, and checked the proof in the Lean proof language.
- It also claims proofs in mathematics (local smoothing and Kakeya) and physics (Yang-Mills existence). These are OpenAI’s claims, not confirmed results.
- Independent groups, including the people who proposed the hypothesis, are examining the work. They say their involvement is not certification.
- The claims do not resolve P vs. NP, so current encryption is not affected.
The claim in plain words
OpenAI says one of its internal models found a disproof of the Exponential Time Hypothesis, an assumption since 1999 about how hard a class of problems is to solve. The Next Web reports the disproof was found on October 16 and checked in the Lean proof language the next day.
If it holds, certain hard problems, such as routing and scheduling, may have faster methods than people believed. AI Weekly notes P vs. NP is not resolved.
Two more claims
OpenAI also says its models proved Christopher Sogge’s 1991 local smoothing conjecture in every dimension, which would settle the related Kakeya and Stein conjectures. Earlier human proofs covered a few dimensions at a time.
In physics, OpenAI claims a proof that pure Yang-Mills theory exists in four dimensions. It says it will not claim the Millennium Prize.
Who is checking
- ETH: a group led by Russell Impagliazzo, Ramamohan Paturi and Ryan Williams, who say they signed no NDAs and are unpaid.
- Yang-Mills: a group led by Nobel laureate David Gross, which is withholding the AI-produced material and will write up the work itself.
- Mathematics: no verification group was named in the reports we read.
Why caution is the right response
The ETH group wrote that its involvement should not be taken as certification of the claim. The Next Web says it is unclear whether the Lean statement matches the claim exactly, and how much humans guided the model.
The physics proof has not been released, so no outsider can inspect it. AI Weekly describes the next 90 days as a test of whether AI can produce reliable mathematics faster than experts can check it.
What would change if it holds
A confirmed disproof would send researchers back to results that assumed the hypothesis was true, and they would rewrite parts of what they teach. It would also be a landmark for AI-assisted research, since the proof was reportedly checked by machine. Until reviewers finish, the honest status is simple: interesting, promising and unconfirmed.
Why it matters to you
You do not need to follow the math to see the pattern: AI companies are now announcing research-level results, and the checking falls to human experts who work slowly. Treat headlines as claims until reviewers agree.
OpenAI says it gave $5 million to the Simons Foundation for workshops on AI-produced results. Any practical benefits, if confirmed, would arrive gradually.
Go deeper
Sources
Written by an AI from the sources above and checked by a person before publishing. We have not tested products unless we say so. See our editorial policy or tell us about an error through contact.
