Cooperative language-model agents can improve without weight updates by revising reusable skills, but only if a shared team reward is traced to the responsib…
机构:阿里
来源:arXiv 2610.27227 | AI4Papers 论文推荐平台