OpenAi claims to have proved that the Exponential Time Hypothesis is false, invalidating many past results that relied on it being true.

Based on 15 sampled comments from this thread. Analyzed 2026-10-27.

The thread treated the ETH result as the real headline: a deterministic, exact counting algorithm with subexponential runtime would overturn a foundational assumption behind a large body of fine-grained complexity work. Readers with relevant background were stunned precisely because the result is not a disguised polynomial-time SAT breakthrough; it leaves the problem superpolynomial, and thus barely shifts informed priors toward P=NP. Several commenters instead saw its importance in closing off a major route for proving hardness and creating a large cleanup-and-rebuild agenda for complexity theory. Much of the enthusiasm centered on a broader pattern: models appear unusually potent in mathematics and formal physics because candidate ideas can be checked cheaply, and formal proof systems supply dense, unambiguous rewards. Commenters contrasted that loop with biology and other experimental sciences, where rejecting a bad hypothesis can require slow, expensive wet-lab work. The mood was exhilarated but unsettled. OpenAI received credit for involving affected experts and delaying release until formalization and review, particularly after earlier attribution criticism. Yet readers worried that frontier labs could become the primary sites of landmark science, with expensive proprietary models and corporate incentives displacing underfunded universities. Some called the capability plainly beyond what scarce human experts could deliver quickly; others demanded evidence on cost, reproducibility, and whether money would have achieved more through direct support for human research.

The sharp dispute was whether this represents genuinely superhuman scientific capacity or an impressive but institutionally troubling deployment of capital and existing human knowledge. Optimists argued that abundant compute can generate scarce expert reasoning at a speed no fixed pool of mathematicians could match; skeptics argued that the comparison is meaningless without costs, crediting prior work, and a serious account of what equivalent research funding might have produced.

• Praise: A formally checkable subexponential algorithm for exact k-CNF counting would demolish ETH while preserving a hard gap from polynomial time, making the breakthrough mathematically substantial rather than hype.
• Praise: Involving domain experts before publication and completing Lean formalization signaled a healthier scientific process after earlier complaints about AI labs absorbing credit for others' work.
• Praise: Fast verification gives AI an unusually strong discovery engine in mathematics and formal physics: wrong conjectures can be discarded cheaply and correct proofs can be mechanically audited.
• Praise: Commenters see a dramatic expansion of scientific search capacity, with frontier systems potentially tackling more open problems than the limited global pool of elite specialists can pursue.
• Concern: The same verification advantage exposes a ceiling in biology, medicine, and materials science, where a bad idea may take days or weeks of wet-lab work to falsify.
• Concern: Private frontier labs may capture paradigm-shifting research while the models that generate it remain inaccessible to universities and the wider scientific community.
• Concern: Excitement over the result is tempered by unresolved questions about compute cost, whether direct human funding could have competed, and whether corporate support survives commercialization.
• Concern: Some readers doubt that OpenAI materially changed its behavior, viewing the new expert-review process as a necessary credibility repair rather than evidence of deeper openness.