EpiAgent: Epistemic Self-Play for LLM Agents
A new arXiv paper introduces EpiAgent, a framework that gives LLM-based agents an explicit, learnable epistemic layer holding beliefs about other agents' private information. The agent's policy is split into tw...