1 paper
Junyu Guo, Yuchen Fang, Shangding Gu +3
As preferences, goals, and facts change, LLM agents must use the current state while earlier versions remain in context. Yet they can answer with an old value of the same variable,…