Exponential Time Hypothesis co-author calls its SI-generated disproof "shocking news"
Prepared by Sintelligent with the help of SI. Published in SI: . Updated: .
SI helped disprove a hypothesis underpinning complexity bounds for many problems. Formal verification is in place, but the practical speedup begins at unattainable problem sizes, and specialists are still examining the result.
Russell Impagliazzo, who proposed the Exponential Time Hypothesis (ETH) with Ramamohan Paturi in 1998, called its disproof "very shocking news". ETH held that the Boolean satisfiability problem 3-SAT could not be solved in subexponential time; hundreds of papers rely on it for conditional complexity bounds. According to OpenAI's account, papers published by OpenAI and Anthropic on October 26 independently propose algorithms with running time 2^o(n); the proofs, produced with substantial SI involvement, have been checked in Lean. Impagliazzo noted that researchers had instead expected stricter versions of the hypothesis to fall first. The practical effect remains limited: according to estimates cited by The New York Times, OpenAI's checked bound improves on brute force only above roughly 10^2389 variables, while Anthropic's version for formulas with five clauses per variable does so above 10^10000. The result does not answer whether P equals NP and does not accelerate NP problems in general. Impagliazzo, Paturi and Ryan Williams are taking part in an independent examination of the work, but stress that this does not constitute certification of the result.
Sources 2