Disentangling Models from Personas in Heterogeneous LLM Simulations

2026-10-05

AI Summary

Existing multi-agent simulations often overlook how differences in underlying models shape interaction dynamics. This study addresses this gap by constructing heterogeneous large language model (LLM) social networks, employing large-scale multi-agent simulation, content mediation analysis, and lexical pattern prediction. Results indicate that base models, rather than role assignments, primarily govern agent engagement, an effect that amplifies as network size increases. Furthermore, network dynamics under heterogeneous model compositions ultimately converge toward base model effects. By challenging the prevalent single-model assumption, this work demonstrates the cross-context predictability of base models and their decisive influence on engagement styles. These findings underscore the necessity of accounting for model heterogeneity in multi-agent research to better understand emergent social behaviors in LLM-driven simulations.

Abstract

Multi-agent simulations with large language models (LLMs) often operate networks of agents with a single base model. This overlooks the inter-model effects which may dominate engagement dynamics in real-world deployments. To show this, we simulate a heterogeneous social network powered by several different base models and show that the amount of engagement an agent receives depends more on its base model than on its assigned persona. The attraction or repulsion effects of a base model strengthen dramatically when more models are added in the mix, suggesting that networks dynamics may converge to base model effects at scale. To help explain this effect, we conduct a series of content-mediating analyses, showing the predictability of base models across contexts as well as the relationship between a model's lexical patterns and an engagement-maximizing style. In light of recent developments in mass multi-agent interaction, this work underscores the relevance of heterogeneous compositions in driving the outcomes of those networks

Problem

• heterogeneous LLM simulations
• multi-agent interactions
• base model effects
• persona disentanglement
• engagement dynamics

Innovation

• Heterogeneous LLM Simulations
• Multi-agent Systems
• Model-Persona Disentanglement
• Engagement Dynamics
• Content-mediating Analysis