AI claims to disprove a key computing hypothesis, stirring scientists

OpenAI and Anthropic each said last week their AI models had disproved the exponential time hypothesis, the idea that some problems can only be solved by trying nearly every answer. Proposed about 25 years ago by Russell Impagliazzo and Ramamohan Paturi, it underpins hundreds of studies. They and MIT's Ryan Williams, a long-time sceptic, are now organising an independent review.