• Sources: primary, discussion, discussion
  • Summary: OpenAI posted a preprint dated 2026-10-25 in its openai/math repository arguing that ETH is false, one of three results announced on 2026-10-26 alongside work on the Kakeya conjecture and quantum chromodynamics. RuntimeWire reports the release includes a Lean formalization and asks computer scientists to examine it.
  • Why it matters: If the result holds, many hardness assumptions behind algorithm lower bounds and fine-grained complexity results would need revisiting.
  • Follow-up: Independent review of the proof and the Lean formalization.