Mathematics of the Future, Science of the Future: Large Language Models and Their Applications

In recent years, the rapid development of large language models (LLMs) such as ChatGPT has given many researchers a strong impression that these systems truly exhibit “intelligence.” In this presentation, we first review the evolution of AI research, explaining how large language models go beyond conventional machine learning by enabling more “general” forms of learning. We then highlight the importance of “sensors” and “mathematical capability” as key factors that allow AI to autonomously carry out scientific tasks such as problem analysis, hypothesis generation, and proofs in fields like mathematics and physics. We also examine how proof assistants can address the issue of hallucinations in LLM outputs, and discuss the role of combinatorial creativity in accelerating interdisciplinary research. Finally, we introduce our “AI Mathematician” agent project, demonstrating how integrating large language models with proof assistants can open new horizons in mathematical sciences.