1 paper
Jiajun Dong, Yutao Hu, Fengrui Fan +3
Large language model (LLM) agents must retain and use cross-step information to act coherently in long-horizon tasks. Existing methods improve memory accessibility, yet action-rele…