Fields Medalist Timothy Gowers: ChatGPT 5.5 Pro Improved an Open Math Bound in Two Hours.

Gowers reports testing ChatGPT 5.5 Pro on open problems from Mel Nathanson's additive number theory paper. The model improved an exponential upper bound to quadratic for sumset diameter. In two hours. Not retrieval of existing solutions. A genuine research contribution. 669 HN points. Gowers warns this has serious implications for PhD training: the "gentle problems" that traditionally serve as entry points for students are now trivially solvable by AI. I don't know what a math PhD looks like in three years, but it doesn't look like this year's.