Tag
This paper investigates formal mechanisms, such as Mediation, to maintain market stability among self-interested LLM agents (DeepSeek-V3) in a simulated marketplace, finding that Mediation enables recovery even under sustained adversarial attacks.
Introduces Agent Bazaar, a multi-agent simulation framework for evaluating economic alignment of LLMs, identifying failure modes like algorithmic instability and Sybil deception, and training a 9B model that outperforms frontier models using targeted reinforcement learning.