OpenAI AI proves Exponential Time Hypothesis false, reopening algorithmic questions

This announcement has generated considerable discussion among mathematicians and participants on Crypto Twitter.

The company's AI models reportedly discovered a method to disprove the ETH, which is a computer-science conjecture originating in 1999. It is crucial to note that the ETH referenced here pertains to theoretical computer science, not the cryptocurrency Ethereum, its price, or co-founder Vitalik Buterin.

Computer scientists Russell Impagliazzo and Ramamohan Paturi, along with Francis Zane, originally proposed ETH. This hypothesis posits that a specific logic puzzle, 3-SAT, cannot be solved significantly faster than by testing every possible combination.

Hundreds of research papers across fields such as graph theory, computational biology, databases, and geometry rely on ETH to suggest that certain problems cannot be solved quickly. OpenAI claims its model has effectively revised this established understanding.

The company stated on its blog that “The refutation of the ETH reopens questions about how much faster these tasks could become,” and that this finding “may help guide the search for better algorithms.” The paper, initially dated October 17 and revised on October 25, details a deterministic algorithm capable of counting solutions to these puzzles in subexponential time.

Independent experts, including Impagliazzo, Paturi, and MIT’s Ryan Williams, are conducting a careful examination of the findings. They issued an open letter stating, “Our involvement should not be taken as certification of the claim,” emphasizing that “a result of this significance requires careful, independent scrutiny and an understanding that others can reconstruct, question, and build upon.”

The paper indicates that the result does not imply that all NP problems can be solved in subexponential time, meaning the P versus NP question remains unresolved. This question concerns whether every problem whose solution is easy to verify is also easy to solve.

In response, Anthropic released a paper on October 26, credited to Claude, titled “A 2^o(n) Time Algorithm for 3-SAT,” which also addresses the ETH problem. The repository notes that the algorithm was found by an internal research Claude model, though the result is described as “purely asymptotic and says nothing about solving formulas in practice.”

This development reflects a broader trend where significant theoretical breakthroughs are now being influenced by AI-driven research efforts, as noted by mathematician Terence Tao regarding the rapid acceleration of AI-powered work on research problems.

Why it matters

This development touches on the fundamental limits of computation, which underpins the theoretical security and scaling assumptions of decentralized systems. While the immediate impact on DeFi or L2s is theoretical, breakthroughs in complexity theory can redefine what is computationally feasible for blockchain operations.