Reinforcement learning is now the standard way to train large language model agents on long-horizon tasks, where dozens of interdependent actions precede a s…
机构:北大
来源:arXiv 2609.12424 | AI4Papers 论文推荐平台