A recent experience with ChatGPT 5.5 Pro
11 hours ago
- #Additive combinatorics
- #AI in mathematics
- #Large language models
- Large language models are advancing rapidly in mathematical problem-solving, now tackling research-level questions and improving upon human solutions.
- ChatGPT 5.5 Pro successfully produced PhD-level combinatorial research in additive number theory, improving bounds on sumsets and restricted sumsets with minimal human input.
- The AI-generated work involved clever use of dissociated sets to achieve polynomial bounds, a non-trivial extension of existing human research.
- This raises questions about the future of mathematical research, including training PhD students, evaluating AI contributions, and where to publish AI-generated results.
- Human mathematicians may need to focus on problems that are hard enough for AI not to solve, or collaborate with AI to push beyond its capabilities.