Large Language Models Through the Lens of Science
Rumen Dangovski, DeepMind
Leuchs-Russell-Auditorium, A.1.500, Staudtstr. 2
Abstract:
Large Language Models (LLMs) hold immense potential for revolutionizing scientific research. In this talk, firstly I present my work on building AI scientists. Then I reflect on how recent LLM advances in multimodality, long-context processing, and agentic applications will accelerate fully automated scientific discovery. Finally, building upon LLM conversational simulation systems with general interfaces that I have developed, I argue that scientists should lead the creation of simulated learning environments for LLMs. Our deep understanding of research positions us uniquely to guide the development of effective learning for frontier LLMs.