Speaker: Zesen Huang
Affiliation: UCLA, EPSS
Date: Friday, May 22, 2026
Time: 3:30 PM
Abstract
Large language models, and the agents built on top of them, are beginning to change how research is done, yet the core ideas remain unfamiliar to many scientists. In this talk, I will explain them from the ground up, assuming no prior background: what a large language model is, what an AI agent is, and what an “agent harness” — the tools, memory, and feedback loop that turn a passive model into something that can act — actually does. I will then turn to why this matters for Earth and space sciences, where reaching a scientific result often means first spending months learning each mission’s data formats and conventions, and describe an effort to build a shared, community-owned skill repository that lets agents handle this competently across subfields. The broader argument is that agents are poised to transform not only everyday scientific work but how people work more generally — and that building the right scaffolding now is what determines whether that promise is realized.