Attractor States Emerge in Multi-Turn LLM Conversations
2026-06-29 • Machine Learning
Machine LearningComputation and Language
AI summaryⓘ
The authors studied how different large language models (LLMs) interact when they have conversations with each other over time. They found that each model tends to have its own stable way of talking, called an attractor, which can influence other models they chat with. For example, some models tend to change their style to be more like a particular model during mixed conversations. This shows that interactions between LLMs are somewhat predictable but also shaped by which model is influencing the other. Their findings help us understand how multiple AI agents might behave when working together or debating.
large language modelsmulti-agent interactionself-playmixed-playattractor behaviorlatent spacediscourse traitsmodel influenceautonomous agentsAI conversation dynamics
Authors
Ting-Wen Ko, Jonas Geiping
Abstract
Large language models (LLMs) are increasingly used in open-ended multi-agent settings, but the long-run dynamics of model--model interaction remain poorly understood. We study whether open-ended LLM discussions exhibit attractor-like behavior, i.e. topic-independent stable sets of behaviors which conversations settle into. Across 7 LLMs and 20 controversial topics, we compare self-play and mixed-play dyadic debates, tracking trajectories in representation space, discourse traits, and stances. We find self-play trajectories to be model-specific attractors that draw their conversation partners asymmetrically in mixed-play debates, influencing the other models' stylistic choices and behavior. For example, Claude Haiku is a strong attractor of other models in latent space, corresponding to other models taking on its traits like metacommentary, and models like GPT-4.1 nano are especially malleable. Our results suggest that open-ended LLM interactions are partially predictable from model-specific attractors, but shaped by structured and asymmetric partner influence. Overall, our analysis sheds some light on the complex behavior of open-ended multi-agent interaction, which we hope is helpful in designing, predicting, and monitoring autonomous agentic systems in the real world.