OpenAI this week announced a batch of results in mathematics, computer science and physics. The most watched claim: its models refuted the Exponential Time Hypothesis, a long-standing conjecture about computational hardness. The pushback followed quickly, with researchers saying OpenAI did not publish enough detail to reproduce the work. The episode highlights a widening tension between frontier labs' announcements and the outside verification academic work depends on.

Source: public reports and official announcements