OpenAI claims its AI disproved a key computer science idea
OpenAI claims an internal AI model has disproved the Exponential Time Hypothesis, a big idea in computer science, Scientific American and The Next Web report. The claim has not been peer-reviewed. Three scientists, two of whom first came up with the idea, will lead an open review.
OpenAI claims an internal AI model has disproved the Exponential Time Hypothesis, or ETH, Scientific American and The Next Web report. The company announced the claim Monday, along with claimed results in math and physics. The claim has not been peer-reviewed.
ETH suggests that a well-known problem called 3-SAT cannot be solved faster than exponential time, Scientific American reports. That means the time needed grows very fast as the problem gets bigger. OpenAI claims its model found a method that is a tiny bit faster. Researchers have used ETH to show that problems in fields such as biology and databases are hard to solve quickly, The Next Web reports.
OpenAI says the model found the claimed disproof on October 16. It says it checked the proof the next day in Lean, a language that verifies proofs by machine, The Next Web reports. The company has posted the write-up and the Lean proof online. A Lean check is not the same as review by the research community.
Russell Impagliazzo and Ramamohan Paturi of UC San Diego, who first proposed ETH with Francis Zane, and Ryan Williams of MIT will lead an open, independent review. "Our involvement should not be taken as certification of the claim," the three wrote in an open letter, The Next Web reports. Williams said the group signed no secrecy deals and will not be paid. OpenAI agreed to release its raw data and fund a conference, according to Williams and OpenAI.
Hours after OpenAI's announcement, rival Anthropic posted its own claimed disproof of ETH, based on work by an internal model, Scientific American reports.
OpenAI also claims proofs of the Kakeya conjecture in every dimension and that Yang-Mills theory exists, part of a million-dollar math prize problem. Scientific American reports the Yang-Mills paper has not been made public. "If the claim is correct, it will change an important part of theoretical computer science," the three ETH reviewers wrote.
What we know: OpenAI claims its AI disproved ETH, and outside experts will review it.
What we don't: Whether the claimed proof holds up under that review.
Where outlets agree Scientific American and The Next Web report OpenAI claims an internal model refuted ETH and that Impagliazzo, Paturi and Williams will examine the claim.
Where they differ The cited reports agree on the key facts so far.