OpenAI says an internal model disproved the Exponential Time Hypothesis, and posts the proof for review
On October 26 OpenAI posted three AI-produced results: a claimed refutation of the Exponential Time Hypothesis via a faster exact k-SAT counting algorithm, a proof of Sogge's local smoothing conjecture, and a 4D Yang-Mills existence claim. The first two ship with Lean formalizations. None is independently confirmed; Impagliazzo, Paturi and Williams are reviewing the ETH claim.
Why it matters
If it holds, an AI system has overturned a core assumption behind many hardness results; for now treat it as an unreviewed claim.