1
PAVE: A Cognitive Architecture for Legitimate Violation in Generative Agent Societies
PAVE架构让AI代理学会在规则社会中“合法越轨”，模拟更真实的人类社会行为。
arXiv:2605.19351v1 Announce Type: cross Abstract: Generative agents based on large language models reproduce believable human behavior in cooperative …