GPT-5.6 Sol reportedly disproves a 30-year-old statistics conjecture in 90 minutes after humans couldn't crack it

A statistics professor at the University of Pennsylvania utilized OpenAI's GPT-5.6 Sol to disprove a central conjecture regarding the Benjamini-Hochberg method in approximately 90 minutes. The predecessor model, GPT-5.5, was unable to find a solution even after 20 hours of attempts. The answer combines known methods in a novel way, raising the larger question: can AI genuinely produce new knowledge, or does it merely recombine what it has already learned? This case sparks a discussion about AI's potential for innovation in research.