What’s Behind OpenAI’s Claim That AI Disproved A Complexity Hypothesis?

🔍 Read the full analysis: What’s Behind OpenAI’s Claim That AI Disproved A Complexity Hypothesis? on ThorstenMeyerAI.com

Age 18–24?Offer from Amazon

Prime made for students and young adults

• Fast, free delivery for dorm and study essentials
• Prime Video and Amazon Music included
• Member-only deals

Try Prime for Young Adults →Free trial for eligible 18–24 year olds

As an affiliate, we earn on qualifying purchases.Hide

OpenAI said October 26 that an unreleased internal AI system found a proof of a subexponential-time algorithm for exactly counting solutions to fixed-width k-SAT, a result that would refute the Exponential Time Hypothesis (ETH) for 3-SAT if the paper’s claims hold. The company has published a preprint and a Lean formalization, but independent expert review has not yet been completed, and the formalization covers a weaker runtime bound than the preprint’s sharpest result.

The 67-page preprint describes a deterministic algorithm that exactly counts satisfying assignments for k-SAT formulas when k is fixed. For 3-SAT, the authors’ claimed running time is subexponential: roughly 2 to the power of n divided by the square root of log n, where n is the number of variables. ETH says that no algorithm can solve 3-SAT in subexponential time. If the proof is correct, the claimed algorithm contradicts that hypothesis. The preprint also says the result refutes the stronger Strong Exponential Time Hypothesis and three related hypotheses.

OpenAI says its system found the disproof on October 16, and that a Lean formalization followed the next day. Lean is a proof assistant: it can check whether a formal proof follows from its stated definitions and rules. The source report counted 3,977 files and about 710,000 lines in the released formalization, while noting that the formal statement itself is short. Machine checking is significant evidence about the proof’s internal logic, but it is not the same as independent mathematical assessment of the assumptions, translation, or relationship between the paper and the formal statement.

There is a specific gap between the paper and the code: according to the source report’s examination of the Lean scope notes, the formalization verifies a slightly weaker runtime bound and does not cover the sharpest bound stated in the preprint. OpenAI says an outside group will explain and verify the computer-science result. The group is led by Russell Impagliazzo and Ramamohan Paturi, two of the three researchers who introduced ETH, alongside Ryan Williams, according to the source material.

A Challenge to SAT Hardness

ETH is used to support lower-bound arguments across theoretical computer science. Researchers often show that a problem is hard by demonstrating that a faster algorithm for it would also yield a faster algorithm for 3-SAT; ETH supplies the assumption that such a speed-up cannot exist. If the new result survives scrutiny, work relying on ETH may need to be revisited, and researchers may need new assumptions or proof techniques.

The claimed algorithm does not make SAT practically easy. Its running time remains superpolynomial, meaning it grows faster than any polynomial in the input size, and the source report says OpenAI itself cautioned that practical improvements would require substantial further research. The implication is about the theoretical boundary of algorithms, not a tool that companies or users can apply to solve large real-world instances tomorrow.

The result also draws attention to a possible research process: an AI system proposes a proof, a formal checker tests a codified version, and specialists examine the mathematics. That process could help researchers explore difficult arguments, but trust depends on what has actually been formalized and on outside review. Here, the distinction between the preprint’s strongest claim and the weaker bound covered in Lean is central.

Products Worth Considering

Samsung 27 Odyssey OLED G5 (G50SF) QHD QD-OLED Gaming Monitor

AmazonView Latest Price

DJI Osmo Mobile 8 Gimbal with Mic Mini Transmitter, Infinity Black

AmazonView Latest Price

Sony WH-1000XM4C Wireless Noise-Cancelling Headphones, Black

AmazonView Latest Price

As an affiliate, we earn on qualifying purchases.

What ETH Says About 3-SAT

In k-SAT, a formula is made of rules, each involving up to k variables or their negations. The task is to decide whether truth values can be assigned so every rule is satisfied; counting versions ask how many assignments work. For n variables, trying all possible assignments takes 2^n steps. Researchers have developed faster-than-brute-force algorithms, but the running time remains exponential in n.

ETH, introduced by Russell Impagliazzo, Ramamohan Paturi and Francis Zane in 2001, asserts that 3-SAT cannot be solved in subexponential time. It is stronger than the statement P ≠ NP: a refutation of ETH would not establish P = NP, nor would it resolve the Clay Mathematics Institute’s still-open P-versus-NP problem. OpenAI’s result is therefore a challenge to a widely used complexity assumption, not a resolution of the central question of whether every efficiently checkable problem is efficiently solvable.

The release is part of a broader set of research claims OpenAI announced on October 26. The company also described work in harmonic analysis and physics, with different outside groups involved in handling those materials. The source report also notes that, after a large release of AI-generated manuscripts on October 6, the repository recorded the withdrawal of three papers over a sign error and revisions to fourteen others. That history is a reason to distinguish formal checking from the separate work of review and correction.

The Proof Still Needs Review

The central open issue is whether independent experts will validate the full argument and its claimed consequences. The source material says the review group is led by researchers associated with ETH’s origins, but it does not report a completed review or a verdict. A Lean checker can verify a formal proof against the encoded definitions and rules; it cannot, on its own, establish that the formalization captures every claim in the paper.

In particular, the published formalization reportedly omits the preprint’s sharpest runtime bound. It is not yet clear from the available material whether that stronger bound can be formalized, whether reviewers will find gaps in the paper, or whether the result will be revised. The preprint’s claims about SETH and three related hypotheses also remain claims pending scrutiny.

There is a separate question about how broadly ETH is regarded as true. OpenAI presents it as a long-standing bet in the field, while the cited survey by Scott Aaronson says there is no consensus. That disagreement does not determine whether this proof is valid, but it matters when describing the result’s reach: refuting a commonly used assumption is not the same as disproving a theorem accepted by every researcher.

External Review Will Test the Claim

The next major milestone is the outside group’s examination of the computer-science work. Its review will need to address the mathematical proof, the connection between the paper and its Lean formalization, and the difference between the weaker formally checked bound and the stronger runtime claim in the preprint. The source material says OpenAI has made the preprint and Lean code public at the group’s request and has agreed to fund a conference on the result.

Until that process produces findings, the careful description is that OpenAI has announced an AI-produced proof that it says refutes ETH, with a related formalization checked by Lean. Whether the full result is correct, how it will affect existing complexity results, and whether any practical algorithmic advances follow remain unresolved.

Key Questions

What does OpenAI say its AI proved?

OpenAI says an internal AI system produced a proof of a deterministic, subexponential-time algorithm for exactly counting solutions to fixed-width k-SAT. For 3-SAT, that would contradict ETH if the proof and its claimed runtime bound are correct.

Does this mean P = NP?

No. A refutation of ETH would not settle P versus NP. The claimed algorithm is still superpolynomial, and P vs. NP remains open.

Has the proof been independently verified?

Not according to the available source material. A Lean formalization has been released and checked by the proof assistant, while an outside group is expected to examine the result. The report says the formalization covers a weaker runtime bound than the paper’s sharpest claim.

Will this make SAT problems easier to solve in practice?

There is no evidence in the source material of an immediate practical speed-up. The claimed running time remains superpolynomial, and OpenAI says practical improvements would require significant further research.

What happens to research that relies on ETH?

If the result is validated, researchers may need to reassess arguments that use ETH to establish hardness. The extent of any changes will depend on expert review and on whether the claimed result is accepted in its full stated form.

Source: ThorstenMeyerAI.com