When a Large Language Model (LLM) is trained with Reinforcement Learning (RL) on multiple tasks, a task's primary reward can gate auxiliary rewards such as a…
机构：CMU
来源：arXiv 2610.36437 | AI4Papers 论文推荐平台