The Epi-LLM Framework: probing LLM behavioral priors through epidemiological agent-based models
Researchers have developed the Epi-LLM framework, which integrates agent-based modeling with large language models (LLMs) to simulate human behavior during epidemics. The framework uses synthetic LLM agents to model societal responses to outbreaks, finding that LLM agents reduced peak infections and achieved significant quarantine compliance. The study also revealed that LLM architecture influences epidemic dynamics, with low-variance models offering better internal validity for behavioral rule testing and high-variance models better representing real-world decision-making. AI
IMPACT Provides a novel simulation environment for pandemic preparedness research, enabling risk-free testing of behavioral interventions.