The Exponential Time Hypothesis Meets AI: Understanding OpenAI’s Results

🔍 Read the full analysis: The Exponential Time Hypothesis Meets AI: Understanding OpenAI’s Results on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

• 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 internal AI model produced a proof refuting the Exponential Time Hypothesis (ETH), a prominent assumption about the difficulty of solving certain logic problems. The company released a preprint and a Lean formalization, but independent expert review has not yet taken place, and the formal check does not cover the paper’s sharpest runtime bound.

The preprint’s central claim is that a deterministic algorithm can exactly count satisfying assignments for any fixed-width k-SAT formula in subexponential time, using polynomial memory. For 3-SAT, that would contradict ETH, which holds that solving such formulas requires exponential time. The paper also says the result refutes the stronger Strong Exponential Time Hypothesis and three related hypotheses.

OpenAI says its model found the disproof on October 16, and the Lean formalization followed on October 17. The preprint is dated October 17 and was revised October 25. The source article reports that the public formalization consists of 3,977 files and about 710,000 lines of code. Lean is a proof assistant: it checks whether a formal proof follows from its definitions and rules. That check is a significant verification step, but it is not the same as independent review of the result’s mathematical significance or the paper’s full claims.

According to the source article’s comparison of the preprint and formal scope notes, Lean checks a slightly weaker runtime bound than the paper states; the formalization does not cover the sharpest bound. 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 ETH’s original authors. OpenAI says collaborators are unpaid, sign no nondisclosure agreements and speak for themselves, and that the company does not seek academic credit.

What an ETH Refutation Would Change

If independent researchers validate the result, it would overturn a widely used assumption in theoretical computer science. ETH has supported proofs that many problems cannot be solved substantially faster than current algorithms allow. A refutation would require researchers to revisit results whose hardness conclusions depend on it, although it would not automatically invalidate every such result; each would need to be checked against the new algorithm and its precise conditions.

The reported runtime is roughly 2 to the power of n divided by the square root of log n. That is subexponential, but still far slower than polynomial time. The source article’s arithmetic says the improvement over exponential time grows slowly as problem size increases. OpenAI itself says practical improvements will require significant further research. The result is therefore a theoretical claim, not evidence that companies can immediately solve large SAT instances more cheaply or quickly.

The development also tests a possible research process: an AI system generates a proof, a proof assistant checks a formalized version, and outside specialists assess the work. The limits matter as much as the headline. The formalization is not as strong as the preprint’s sharpest runtime claim, and external experts have not yet signed off. Trust in an AI-produced result depends on the public evidence and scrutiny, not solely on the fact that software accepted a formal proof.

Popular right now

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

AmazonView Latest Price

Breville Bambino Espresso Machine, Brushed Stainless Steel

AmazonView Latest Price

Dyson Ball Animal 2 Plus Corded Vacuum

AmazonView Latest Price

As an affiliate, we earn on qualifying purchases.

From SAT Algorithms to ETH

In k-SAT, each rule is a logical clause involving at most k variables, and the task is to determine whether there is an assignment that satisfies all clauses. With n variables, trying every possible assignment takes 2^n checks. Researchers have improved on brute force by reducing the exponential base: the source article cites Schöning’s 1999 algorithm at about (4/3)^n and later PPSZ refinements at about 1.307^n.

ETH, proposed by Russell Impagliazzo, Ramamohan Paturi and Francis Zane in 2001, goes beyond saying that no polynomial-time algorithm is known. It asserts that 3-SAT cannot be solved in subexponential time. The source article notes that views on whether ETH is true are not unanimous: Scott Aaronson’s 2017 survey said there was no field-wide consensus. Refuting ETH would also not settle P versus NP, a separate open question about whether every problem whose answer can be checked efficiently can also be solved efficiently.

OpenAI presented the ETH work alongside two other claimed results: a proof concerning Sogge’s local smoothing conjecture in harmonic analysis, and work in mathematical physics concerning quantum chromodynamics and a grand unified completion of the Standard Model. The company said different outside groups would guide the release of each project. The physics materials were being held back, according to OpenAI’s account.

Formal Checks and Review Gaps

The central open issue is whether independent specialists agree that the preprint’s argument is correct and that its conclusions follow as stated. The announced review group has not yet completed its examination, and the source material provides no review outcome. The formalization’s weaker runtime bound also leaves a gap between what Lean checked and the strongest bound asserted in the paper.

It is not yet clear whether reviewers will validate the full preprint, the weaker formally checked result, or both after revisions. Nor does the announcement establish how much the algorithm improves performance on real-world instances. Those questions require mathematical scrutiny and, separately, evidence from implementation and testing.

Independent Review Will Test the Claim

The next milestone is the external group’s examination of the computer-science preprint and its Lean code. OpenAI says the group will explain and verify the result; the source material does not give a completion date. Researchers’ findings could clarify which runtime bounds are established, whether the formal statement matches the paper, and what changes or corrections may be needed.

Until that review is reported, the result should be described as an AI-generated proof claim with a machine-checked formalization, not as an independently confirmed overturning of ETH. Any later practical impact will depend on further work to develop, analyze and test algorithms for relevant problems.

Key Questions

What is OpenAI claiming?

OpenAI says an internal AI model produced a proof that a deterministic algorithm can exactly count solutions to fixed-width k-SAT in subexponential time. For 3-SAT, the claim contradicts ETH.

Has the proof been independently confirmed?

Not yet, according to the supplied source material. A Lean formalization has been released, and an outside group is expected to review the result.

Does this solve P versus NP?

No. ETH is a stronger, separate hypothesis. Even if the reported result is confirmed, P versus NP remains open.

Does the result make SAT problems easy to solve in practice?

That has not been shown. The reported algorithm remains slower than polynomial time, and OpenAI says practical improvements will require significant further research.

What did Lean verify?

Lean checked a formalized proof with a slightly weaker runtime bound than the sharpest bound in the preprint. The formalization therefore does not cover the paper’s strongest stated runtime claim.

Source: ThorstenMeyerAI.com