It is common to train Large Language Models (LLMs) for agentic tasks with Reinforcement Learning (RL) from scalar rewards. However, this throws away the fine…
机构:NYU
来源:arXiv 2611.16409 | AI4Papers 论文推荐平台