Fields Medalist Timothy Gowers reports that ChatGPT 5.5 Pro improved an exponential bound to a polynomial one in under an hour. An MIT researcher described the model's core insight as completely original. This result suggests a new baseline for mathematical research where human value lies in solving what LLMs cannot.