OpenAI, Anthropic Refute Exponential Time Hypothesis — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI released solutions to three blockbuster math problems, one of them a rejection of the Exponential Time Hypothesis, a durable assumption about what is hard to compute that many papers rely on.
- Anthropic released its own ETH refutation hours after its mathematician Levent Alpöge suggested OpenAI's solutions were based on pre-existing work.
- OpenAI split its release by discipline: mathematicians got the work dropped on GitHub, computer scientists formed a working group that asked for the proofs to be shared publicly and for OpenAI to fund a conference, and physicists formed a working group that will decide how their result is released.
- Ryan Williams, the MIT professor organizing the computer science response, called OpenAI's earlier October proof dump a "DDOS attack on mathematicians" but agreed to collaborate, describing parts of the new proof as "alien-looking."
- Russell Impagliazzo, one of the UC San Diego computer scientists who first published ETH in 1999, joined Williams and Ramamohan Paturi in an open letter calling for scientific feedback and discussion of the norms around AI contributions to the field.
- Dan Roberts, OpenAI's science lead, was asked by Williams whether the faster algorithm breaks cryptographic systems; OpenAI checked and was reasonably confident it did not imply any breaks.
- P versus NP — Williams said every computer scientist should put more weight on the plausibility that P = NP, while cautioning that OpenAI's algorithm is probably not the best obtainable and that ETH-based work from the past decade will need revisiting if the claim holds.
Why it matters: Computer scientists are the immediate audience: Williams said every one of them should put more weight on P = NP, while papers built on ETH across the past decade face revision. OpenAI's own crypto check found no immediate breaks, so the pressure lands on peer review rather than on deployed security.
Ask SkimNews
