1 paper
Haoran Liu, Yuwei Zhang, Xiyao Li +2
Reinforcement learning typically improves multi-turn agent capabilities through the terminal outcome of the trajectories, which makes it difficult to determine credit assignments f…