After AI Claims to Topple a Computer Science Pillar, Researchers Weigh Years of Work Left 'Irrelevant'

The claim that AI has disproved the exponential time hypothesis (ETH) is now turning into a very human story. A day after OpenAI and Anthropic each said their models had knocked down the 27-year-old assumption, computer scientists are describing what it means for their own work, and the picture is equal parts loss and excitement.

In a report published Thursday, Oct. 29, WIRED spoke with researchers who say the result, if it holds, could leave years of their papers “dead,” “pointless” or in need of a rewrite.

What is the AI breakthrough WIRED reported?

Both OpenAI and Anthropic said this week that their models had found a way to solve a classic logic puzzle, satisfiability, faster than the ETH says is possible. The hypothesis held that the effort to solve such problems grows exponentially as they get bigger, and a large body of research on algorithms and complexity has been built on that premise, WIRED reported.

Both companies have posted their proofs on GitHub, in plain-language form and in Lean, a programming language that machines can use to check proofs. Researchers have not yet independently verified the proofs, or even had time to fully understand them, WIRED noted. Science reported that Russell Impagliazzo, who co-proposed the ETH in 1999 and is helping review the work, said, “As far as I’m concerned, this is not 100% certain.”

How AI impacts theoretical computer science research

The researchers WIRED interviewed described the fallout in vivid terms:

• Peter Jonsson of Linköping University is due to have work presented in two weeks at the IEEE Symposium on Foundations of Computer Science, one of the field’s top conferences. He said much of that work is now “dead.” He expects many talks there to start with, “ETH is false, half of my paper does not work.”
• Daniel Lokshtanov, a UC Santa Barbara professor, estimated that roughly 50 of his published papers could become “pointless” or need revisiting if the result is confirmed. “It’s just a giant bomb dropped into the middle of the field,” he said. “It’s really cool and terrible at the same time.”
• Thore Husfeldt, who heads the algorithms research group at the IT University of Copenhagen, said a problem he spent most of this year on now looks “all irrelevant.”

That fits what Science reported earlier in the week: Anthropic counted more than 800 papers that prove results only by assuming the ETH or a stronger version of it.

“Overexcited bunnies”: why some researchers are thrilled

Not everyone is mourning. Husfeldt called the finding “fantastic.” He compared math to “stumbling about in a completely dark room” and said the result was like someone lighting up a whole subfield, leaving researchers to “run around like overexcited bunnies” exploring it, according to WIRED.

Victor Lagerkvist, an associate professor at Linköping, told the magazine he plans to go back to problems he had given up on because he had assumed the ETH was true. Jonsson, despite the hit to his paper, said a clear answer on the ETH was something he could only “dream of” before. “These are exciting times,” he said.

The strain on the field

Researchers also described a heavier toll. Husfeldt told WIRED that the past few months have been “really tough,” with AI results regularly wiping out ongoing projects, and that his department has considered hiring an occupational psychologist because some colleagues are “really depressed and really sad.”

Academics also complained about the sheer volume of AI-generated material. WIRED reported that OpenAI earlier this month posted 722 manuscripts on 400 problems to GitHub, some with errors, frustrating researchers who struggled to digest them. Academics told the magazine that OpenAI’s roughly 60-page ETH paper appears better written than the companies’ past releases.

Does this affect encryption?

Not directly, according to the reporting. Online security depends on certain math problems taking impractically long to solve. Lokshtanov said this result doesn’t directly affect cryptography, but “it should definitely scare some cryptosystem designers,” WIRED reported.

What happens next

OpenAI has asked the researchers who originally proposed the hypothesis, plus an MIT computer scientist, to organize a review of the proofs, and has committed to fund a conference on the topic, WIRED reported. Science said the plan includes a fully human-readable proof and a virtual review conference before the end of the year.

FAQ

Earlier coverage: AI Claims to Break the Exponential Time Hypothesis, Shaking Theoretical Computer Science (Oct 29)