OpenAI says its newest AI system has refuted the exponential time hypothesis, a conjecture proposed in 1998 that certain computational problems cannot be solved quickly. The result does not settle the famous P versus NP problem, which asks whether problems whose answers are easy to check are also easy to solve. But it challenges a closely related assumption: that finding solutions to some hard problems must take far longer than verifying them. The idea is illustrated by Sudoku: checking a completed grid is straightforward, while solving a difficult one can be laborious. The AI’s proof suggests that unexpectedly efficient “shortcuts” for solving such problems may exist, even for enormous puzzles. Anthropic separately announced that it had also disproved the hypothesis, amid a wave of AI-generated results in mathematics and science.
The finding is theoretical, with no immediate practical applications, and experts say translating such results into working technology could take years or decades. Still, if the shortcuts can eventually be implemented, they could help computers design optimized circuits and chips, discover medicines by modeling proteins, or solve problems relevant to encryption. The most consequential possibility is recursive self-improvement: AI systems might use these methods to design more capable systems efficiently. That prospect makes the result notable not only because a machine produced a surprising mathematical advance, but because it may point toward ways AI could help build its own successors. The underlying P versus NP question remains open, and the conjecture’s disproof does not show that today’s computers can suddenly solve currently intractable problems.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.