Lei Ge is a PhD Candidate at Imperial College London, where her research focuses on large language models for science, interpretable AI, and agentic AI. She is also a part-time Machine Learning Engineer at Polaron, developing LLM-agent systems for scientific and industrial applications. Her work explores how foundation models can support scientific discovery and optimization in materials research, while also examining how their behaviour can be understood, interpreted, and trusted. More broadly, she is interested in both the technical challenges of AI for science and the evolving role of scientific practice in the age of intelligent machines.

Ms. Ge’s talk is titled ‘From Trust to Action: Large Language Models for Scientific Discovery and Decision Support’ and handles the question ‘Can LLMs be trusted in scientific settings?’

This talk explores that question by asking what it would take for large language models to move from impressive demonstrations to reliable components of scientific practice. The focus is not only on what these models can do, but on what their behaviour reveals: what they know, how they respond to uncertainty, and when their performance can be trusted in constrained optimization and real scientific workflows. More broadly, the talk considers how intelligent systems may begin to reshape the relationship between scientific knowledge, experimentation, and decision-making across research and industry.