Anthropic's Claude finds 2^o(n) 3-SAT algorithm, refuting the Exponential Time Hypothesis
rickasaurus · x · 2026-10-27
- A paper on Hexagon, labeled "primarily AI-generated," reports that an internal Anthropic research Claude discovered, with zero human input and 14M output tokens, a deterministic 3-SAT algorithm running in 2^{o(n)} time.
- The algorithm also solves the harder #3-SAT (counting satisfying assignments), which would refute the Exponential Time Hypothesis (ETH), a central conjecture in complexity theory.
- The full result has been formalized in Lean 4; found Oct 16, with Anthropic researchers only polishing the writeup.
- An internal OpenAI model independently claims a similar proof, discovered concurrently.
- Sharer rickasaurus concludes he's now in the "humans were just really bad at this" camp.
More from AGI Musings
- Sabine Hossenfelder: AI will mine hidden scientific discoveries before it replaces truck drivers — skdh · 2026-10-27
- Brynjolfsson, Agrawal and Korinek outline nine grand challenges for transformative AI economics — soumitrashukla9 · 2026-10-27
- AI researcher argues superintelligence unpredictability means we can't assume it stays human-friendly — danfaggella · 2026-10-27
- Language Is a Lossy Copy of Thought: Why Brain-Interfaces and AI May Bypass Words — r0ck3t23 · 2026-10-27
- Friction is the new distribution: AI erodes old moats, complexity becomes defensibility — vaibhavbetter · 2026-10-27
- OpenAI's 2023 labor-impact paper revisited: mathematicians top the exposure list — skdh · 2026-10-27