Social World Model Framework Predicts Belief Shifts from Events Using LLMs
June 9, 2026
Social World Model (SWM) learns state-transition functions for social beliefs by mining temporal patterns in social data and optimizing an evidence lower bound, requiring no explicit human annotations linking events to belief shifts. The accompanying SWM-bench benchmark evaluates how well models predict belief evolution following major social events.
HOW THIS AFFECTS YOU
●
researcherSWM-bench provides a structured evaluation for LLM social reasoning that goes beyond static commonsense benchmarks, useful for studying temporal belief dynamics.