Belief Coevolution in a Social Network of Generalist and Specialist Large Language Models: A Controlled Study
The study used a CoevolveSim framework to track how beliefs spread among different types of large language models (LLMs) in a network. It found that the diversity of LLMs, specifically introducing specialist agents, is the main driver of collective belief change, far more important than just assigning social roles. This heterogeneity leads to stronger influence asymmetries and larger shifts in group consensus.