OpenAI announces the falsification of the important theory "Exponential Time Hypothesis (ETH)" that has been used in the computer science field for over two decades.

October 27 - OpenAI recently announced that its AI model has "cracked" ETH—this news quickly sparked heated discussions in the tech and crypto circles. But it needs to be specially clarified that the ETH here is not Ethereum, widely known in the cryptocurrency field, but an important conjecture that has existed in computer science for more than twenty years: the Exponential Time Hypothesis (ETH).

The core question of the Exponential Time Hypothesis is: how hard are those problems that seem "especially hard to compute"?

Take the "SAT problem" (Boolean satisfiability problem), which some people may have heard of, as an example. It asks: given a set of logical conditions, can you find an assignment that makes all conditions hold at the same time? As the scale of the problem grows, the best known method at present can basically only try possible answers one by one, and the time cost grows exponentially with scale—what we often call "exponential time." ETH further asserts that such problems fundamentally cannot be substantially accelerated, and that there is no truly "sub-exponential time" efficient algorithm.

Since this hypothesis was proposed in 1999, it has been an important cornerstone of theoretical computer science. Hundreds of papers rely on it to prove that "such-and-such problem is hard" and that "such-and-such algorithm is already close to optimal." If ETH were overturned, it would mean that many problems we thought were "almost impossible to solve faster" might in theory have better solutions.

OpenAI this time claims that its model has found a deterministic algorithm that can exactly count the number of solutions to the k-SAT problem in sub-exponential time, and has formally verified it with Lean. If this result ultimately withstands expert review, it will directly shake the Exponential Time Hypothesis.