LLMs as Human-Centered World Models
This project explores LLMs as data-driven world models for simulating how communities and infrastructure may experience a disaster before it occurs.
Technical innovation
The research integrates heterogeneous pre-event evidence with language-model reasoning to represent human behavior, infrastructure disruption, and environmental conditions in a shared simulation framework.
Practical insight
Pre-event scenarios become more useful when they represent people as more than aggregate demand. Human-centered simulation can reveal differentiated impacts, overlooked vulnerabilities, and intervention opportunities before decisions become urgent.