Frontier AI labs say their models refuted the Exponential Time Hypothesis. What does that mean?
OpenAI and Anthropic say their AI systems produced proofs refuting the Exponential Time Hypothesis (ETH), a long-standing assumption that certain computational problems cannot be solved quickly as they grow. If the result holds up, it could change how computer scientists understand the limits of efficient computation, though researchers are still examining the proof.
MIT computer scientist Ryan Williams is coordinating a response with researchers Russell Impagliazzo and Ramamohan Paturi, who introduced ETH in 1999. Williams described parts of OpenAI’s proof as unusual and called for community feedback and discussion of standards for AI contributions; OpenAI has also proposed funding a conference. The article frames ETH within the wider P versus NP question, which asks whether problems whose answers can be checked quickly can also be solved quickly.
In short
- OpenAI and Anthropic say their models refuted ETH.
- Researchers are organising scrutiny of the proofs.
- The claim could affect assumptions about computational difficulty.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
If the proofs withstand scrutiny, they could overturn a foundational assumption and reveal that some problems thought to require exponential time have faster solutions. AI systems may help researchers explore difficult mathematical arguments, and open review, community discussion and a dedicated conference could help assess the claims fairly.
The case against
A claim this consequential needs careful independent verification: unusual proof steps may conceal gaps, and the researchers are still examining the work. The episode also raises legitimate questions about how AI-generated mathematics should be checked and credited, so the community should avoid treating the claims as established before the proofs meet accepted standards.
Explained simply
Computer scientists study how the time needed to solve a problem grows as the problem gets larger. The Exponential Time Hypothesis, or ETH, is an assumption that some problems need vastly more time as they grow, rather than being solvable efficiently. It helps researchers compare the difficulty of problems and reason about the limits of computing.
ETH is linked to P versus NP, a famous open question about whether every problem whose answer can be checked quickly can also be solved quickly. Ryan Williams, Russell Impagliazzo and Ramamohan Paturi are among the computer scientists associated with studying these questions; Impagliazzo and Paturi introduced ETH in 1999. Claims that ETH can be disproved matter because they could reshape ideas about which computational problems can be solved efficiently.
AI Research Science Technology