OpenAI’s AI Findings Challenge A Hypothesis—But What’s Established?

🔍 Read the full analysis: OpenAI’s AI Findings Challenge A Hypothesis—But What’s Established? on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on monitors, keyboards and dev gear

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

OpenAI said October 26 that an unreleased AI model had produced a proof of a subexponential-time algorithm for exactly counting solutions to fixed-width SAT formulas, a result that would contradict the Exponential Time Hypothesis (ETH) for 3-SAT. The proof has a public Lean formalization, but independent expert review is pending, and the formalized result checks a weaker runtime bound than the sharpest claim in the preprint.

The preprint claims a deterministic algorithm that counts the satisfying assignments of any k-SAT formula, for fixed k, in subexponential time and with polynomial memory. For 3-SAT, that would contradict ETH, which says there is no algorithm solving the problem in subexponential time. The authors also say the result refutes the stronger Strong Exponential Time Hypothesis and three related hypotheses. Those are claims in the preprint, not conclusions that have received independent sign-off.

OpenAI says its system found the disproof on October 16, and that a Lean formalization followed the next day. The preprint is dated October 17 and was revised October 25. The source report says the published formalization contains 3,977 files and about 710,000 lines of code, while the theorem’s formal statement is short. Lean can check that a formal proof follows from its definitions and rules; that check does not, by itself, establish that the work has passed outside expert scrutiny or that the formalized theorem matches every strongest claim in the paper.

That distinction matters here: according to the report, Lean verifies a slightly weaker runtime bound than the preprint’s sharpest one, which is not covered by the formalization. A review group led by Russell Impagliazzo, Ramamohan Paturi and Ryan Williams is expected to examine the computer science result. Impagliazzo and Paturi were among the original authors of ETH. OpenAI says the preprint and code were made public at the group’s request.

What an ETH Refutation Would Change

If the proof survives scrutiny, it would overturn a widely used working assumption in theoretical computer science. ETH has supported many conditional lower bounds: results showing that certain problems are unlikely to have substantially faster algorithms, provided ETH holds. A genuine refutation would require researchers to revisit conclusions that depend on that assumption. It would not automatically invalidate every such result, since each implication and its conditions would need to be checked.

The claimed algorithm is subexponential, not polynomial-time. The source report describes its running time as roughly 2 to the power of n divided by the square root of log n. That remains far slower than polynomial time as problem size grows, and says little on its own about performance on real-world instances. OpenAI itself cautioned that practical improvements would require significant further research. No immediate speed-up for software or industry follows from the announcement.

The development also tests a proposed research process: an AI system generates a mathematical result, proof-assistant software checks a formal version, and human specialists examine the result. A successful formal check can catch gaps in a proof encoded in the system, but the gap between the paper’s strongest theorem and the weaker formalized bound shows why the precise scope of that check needs to be stated. The case may shape how future AI-produced research is released and assessed.

Products Worth Considering

Samsung 27-Inch Odyssey QD-OLED G8 4K 240Hz Gaming Monitor

AmazonView Latest Price

Apple iPad Air (4th Generation, 10.9-inch, Wi-Fi, 256GB) – Space Gray

AmazonView Latest Price

Amazon Kindle Paperwhite (12th Generation, 2024), 16 GB, 7-inch Display, Jade

AmazonView Latest Price

As an affiliate, we earn on qualifying purchases.

ETH, SAT and the P–NP Distinction

SAT asks whether there is an assignment of true or false values that makes every logical clause in a formula true. In k-SAT, each clause contains at most k variables or their negations. A formula with n variables has 2^n possible assignments, though researchers have developed algorithms faster than checking every assignment. The question behind ETH is whether 3-SAT can be solved in time that is subexponential in n.

ETH was proposed in 2001 by Russell Impagliazzo, Ramamohan Paturi and Francis Zane. It is stronger than the open P-versus-NP question. P versus NP asks whether every problem whose solutions can be checked efficiently can also be solved efficiently; ETH makes a more specific claim about the difficulty of 3-SAT. So even if the announced result is confirmed, it would not prove P = NP. The P-versus-NP problem remains open.

OpenAI’s October 26 announcement also described separate mathematics and physics manuscripts, each handled with outside academic groups. The ETH work is the result for which OpenAI says the preprint and Lean code were released. The source report also notes that an earlier repository release had manuscripts withdrawn or revised after errors were found, a reminder that publication and formalization are not substitutes for review.

The Proof’s Scope and Review

Independent verification has not been completed, according to the source report. The review group’s examination is the next test of whether the paper’s argument is sound and whether its claims follow from the stated assumptions. The public Lean check is meaningful evidence about the proof it encodes, but it covers a weaker runtime bound than the paper’s sharpest result.

The available material does not establish when the reviewers will report, what changes they may request, or whether they will endorse the full preprint. It is also unclear what practical performance the claimed algorithm could achieve, if any, on problems encountered outside theoretical analysis. The claims about refuting SETH and related hypotheses likewise await scrutiny.

Reviewers Take Up the Result

The group led by Impagliazzo, Paturi and Williams is expected to explain and examine the computer science result. The key milestones will be whether that review validates the paper’s main theorem, how it addresses the difference between the preprint and the Lean-checked bound, and whether the formalization is expanded to cover the stronger claim. Until then, the result should be described as an announced, machine-checked claim awaiting independent review—not as a settled change to computer science.

Key Questions

What does OpenAI say its AI proved?

OpenAI says the model produced a proof of a deterministic subexponential-time algorithm for exactly counting solutions to fixed-width k-SAT formulas. For 3-SAT, that would contradict ETH. Independent review is still pending.

Does this solve P versus NP?

No. ETH is a stronger, more specific hypothesis about the difficulty of 3-SAT. The announced algorithm is still superpolynomial, and P versus NP remains open.

Has Lean checked the entire result?

Not the preprint’s sharpest runtime bound, according to the source report. Lean checks a slightly weaker bound in the public formalization. Outside experts have not yet signed off on the result.

Will SAT problems become easier to solve now?

There is no confirmed immediate practical effect. The claimed algorithm is a theoretical result, and OpenAI said practical improvements would require significant further research.

What happens next?

An outside group led by Russell Impagliazzo, Ramamohan Paturi and Ryan Williams is expected to examine the computer science result. Its assessment and any changes to the paper or formalization remain pending.

Source: ThorstenMeyerAI.com