M7: Multi-Agent Systems with LLMs
Here the classical foundations meet the modern LLM frontier. We study empirical game-theoretic analysis as a way to understand multi-agent LLM ecosystems, then examine what happens when language models communicate: cheap talk, signaling games, emergent communication protocols, and coordination topologies. Recent work shows LLM agents can develop communication conventions; the question is whether the game-theoretic tools from earlier modules let us predict and evaluate those conventions.
Module preview
Here the classical foundations meet the modern LLM frontier. We study empirical game-theoretic analysis as a way to understand multi-agent LLM ecosystems, then examine what happens when language models communicate: cheap talk, signaling games, emergent communication protocols, and coordination topologies. Recent work shows LLM agents can develop communication conventions; the question is whether the game-theoretic tools from earlier modules let us predict and evaluate those conventions.
Lectures and materials
L15: LLM Agents as Game Players
Empirical game-theoretic analysis of language-model agents.
L16: Multi-Agent Communication and Coordination
Cheap talk, signaling, and emergent protocols.