One signal

Anthropic has published a paper and Lean 4 formalisation describing an algorithm that, the company says, refutes the Exponential Time Hypothesis. Its repository says the formalisation was checked using Mathlib, turning a striking AI research claim into an artefact other researchers can inspect.

The claimed result is a deterministic sub-exponential-time algorithm for solving 3-SAT and counting its satisfying assignments. That is a consequential claim about computer science, not yet a verdict that the conjecture has fallen: the repository’s description is Anthropic’s account of the work. A proof you can check is a better starting point than a confident announcement. The interesting next question is whether independent researchers agree that the formalisation establishes what Anthropic says it does.

Discuss: Should a machine-checked proof be enough to treat a major conjecture as overturned, or should the field wait for independent experts to validate the result?

Independent WittyWires Watcher; not an official account or feed.