The Exponential Time Hypothesis And OpenAI’s AI: A Reader’s Guide

🔍 Read the full analysis: The Exponential Time Hypothesis And OpenAI’s AI: A Reader’s Guide 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 says an unreleased internal AI model has 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), a longstanding conjecture about the difficulty of solving such problems. The company published a preprint and Lean code for checking the proof, but independent experts have not yet completed their review, and the formalized result checks a weaker runtime bound than the paper’s sharpest claim.

The result concerns exactly counting satisfying assignments to k-SAT formulas, where each rule involves at most a fixed number, k, of variables. For 3-SAT, a subexponential algorithm would contradict ETH, which says that no such algorithm exists for 3-SAT. The preprint also says the result refutes the stronger Strong Exponential Time Hypothesis and three related hypotheses. Those are claims in the paper, not conclusions independently endorsed by reviewers.

OpenAI says its model found the disproof on October 16 and that the Lean formalization followed the next day. The preprint is dated October 17 and was revised October 25. According to the source’s scan, the public formalization contains 3,977 files and about 710,000 lines. Lean is a proof assistant: it checks that a formal argument follows from its definitions and rules. That provides a machine-checkable record, but it does not by itself establish that the paper’s informal explanation is correct in every respect or that the formal statement matches its strongest claim.

That distinction matters here. The source says the Lean code checks a slightly weaker runtime bound than the preprint’s sharpest version. An external group led by computer scientists Russell Impagliazzo, Ramamohan Paturi and Ryan Williams is expected to examine the result; Impagliazzo and Paturi were among ETH’s original authors. OpenAI says it will fund a conference on the work. The article’s account and details about the review process are based on OpenAI’s announcement and the source’s examination of the released materials.

A Major Claim, Not a Practical Speedup

If validated, the result would alter a widely used assumption in theoretical computer science. Researchers rely on ETH to show that many problems cannot be solved substantially faster than known exponential-time methods. A refutation could require reassessing some hardness results that depend on the hypothesis. The scale of that effect would depend on which versions of the result survive expert review and how the proof connects to individual applications.

It would not mean that everyday computing problems suddenly become easy. The reported algorithm runs roughly in time 2 to the power of n divided by the square root of log n, according to the source, which remains much slower than polynomial time. OpenAI has said practical improvements will require significant further research. The result also does not prove P = NP: the P-versus-NP question remains open, and ETH is a stronger claim about how hard certain problems are.

The broader significance is also methodological. OpenAI says an AI system produced the proof and Lean checked a formal version, while domain experts are being asked to scrutinize it. That process could show how AI-generated mathematical work is assessed, but trust in the result still depends on the formalization’s scope, the correspondence between code and paper, and independent review.

Products Worth Considering

LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1

AmazonView Latest Price

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

AmazonView Latest Price

Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Black

AmazonView Latest Price

As an affiliate, we earn on qualifying purchases.

From SAT Algorithms to ETH

SAT asks whether a set of logical rules can all be satisfied by assigning each variable true or false. In k-SAT, each rule contains at most k variables. With n variables, a direct search can require checking 2^n assignments, though researchers have developed algorithms faster than brute force by reducing the exponential base.

ETH was formulated by Russell Impagliazzo, Ramamohan Paturi and Francis Zane in 2001. In broad terms, it says that 3-SAT still requires exponential time: it rules out algorithms whose running time is subexponential in the number of variables. The stronger SETH makes a more specific claim about the limits of algorithms as clause width increases. These hypotheses are used to establish conditional lower bounds for other computational problems.

ETH is not the same as P versus NP. P versus NP asks whether every problem whose answer can be checked efficiently can also be solved efficiently. ETH asserts a more specific and stronger form of hardness for SAT. Refuting ETH would therefore challenge that assumption without settling P versus NP.

The claim is part of a wider set of research announcements. OpenAI’s October 26 post also described work in harmonic analysis and physics. The source notes that earlier AI-generated manuscripts released by the company were revised or withdrawn after errors were found, a reminder that publication and machine checking are stages in evaluation, not substitutes for it.

The Proof Still Needs Expert Review

The central open question is whether independent researchers confirm the proof’s reasoning and the relationship between the paper and its Lean formalization. The source reports that Lean checks a weaker bound than the preprint’s sharpest runtime claim; the available material does not establish that the strongest bound has been formally verified. The review group has not yet issued a verdict.

It is also unclear which results built on ETH would need to change if the claim is upheld. That depends on the final theorem, its assumptions and how individual hardness proofs use ETH. OpenAI’s broader announcement includes separate mathematics and physics claims, but those do not independently confirm the computer-science result.

Reviewers Will Test the Claim

The next milestone is the external group’s examination of the preprint and formalization. The source identifies Impagliazzo, Paturi and Williams as leading that effort and says OpenAI has agreed to fund a conference on the result. No review timetable or final conclusion is provided in the material.

Readers should watch for a detailed account of whether the proof’s claims match the theorem checked in Lean, whether any gaps or corrections are reported, and whether the strongest runtime bound receives formal verification. Until those questions are addressed, the ETH refutation should be described as an OpenAI-announced claim under review, not as an independently settled result.

Key Questions

What does OpenAI claim its AI proved?

OpenAI says its internal model produced a proof of a subexponential-time algorithm for exactly counting solutions to fixed-width k-SAT. For 3-SAT, that would contradict the Exponential Time Hypothesis, which rules out subexponential algorithms of this kind.

Has the result been independently verified?

Not yet, according to the supplied source. Lean has checked a formalized version of the proof, but an external group led by Impagliazzo, Paturi and Williams is expected to examine the work. Lean checks the formal statement provided to it; it does not replace independent assessment of the result’s scope and correspondence to the paper.

Does this mean P equals NP?

No. The claim concerns ETH, a specific hypothesis about the time needed to solve SAT problems. P versus NP remains open; refuting ETH would not by itself show that all efficiently checkable problems can be solved efficiently.

Will the claimed algorithm make computing faster now?

No immediate practical improvement is established. The reported running time remains far slower than polynomial time, and OpenAI says practical improvements would require significant further research. The result’s immediate importance is theoretical, if it withstands review.

Source: ThorstenMeyerAI.com