Mathematician Tests ChatGPT 5.5 Pro on PhD-Level Research
Original: A recent experience with ChatGPT 5.5 Pro
Why This Matters
Demonstrates AI's growing capability to conduct original mathematical research at graduate level.
Cambridge mathematician Timothy Gowers reports that ChatGPT 5.5 Pro produced PhD-level mathematical research within an hour, solving open problems from additive number theory with minimal human input, marking a significant advancement in AI mathematical capabilities.
Timothy Gowers, a prominent Cambridge mathematician, tested ChatGPT 5.5 Pro on open research problems in additive number theory from a paper by Mel Nathanson. The AI system successfully solved PhD-level problems within approximately one hour with minimal mathematical guidance from Gowers. Gowers notes that LLMs have progressed beyond simply finding existing solutions in literature to potentially discovering arguments that human mathematicians have missed. He experimented with problems involving sumsets and combinatorial parameters, areas where authors typically don't spend extensive time on each question due to volume. The mathematician observes that the bar for research problems has been raised - they now need to be difficult enough that LLMs cannot solve them. This development suggests traditional sources of problems for beginning researchers may no longer be sufficient.