LLM-assistedScientific Experimentation? Transforming Science with LLMs

Abstract

This talk explores the emerging role of large language models (LLMs) in scientific experimentation, focusing on two complementary roles: LLMs as scientific programmers and LLMs as machine-learning experiment designers. Through examples including ScienceAgentBench, SciCode, and AutoML-GPT, it discusses how LLMs can support computational scientific tasks, data-driven discovery, and increasingly automated experimental workflows, while highlighting current limitations and the path toward more autonomous scientific experimentation. The presentation draws on the survey “Transforming Science with Large Language Models: A Survey on AI-assisted Scientific Discovery, Experimentation, Content Generation, and Evaluation,” accepted in ACM Computing Surveys (2026). Survey paper: https://arxiv.org/abs/2502.05151