https://www.cheasim.com/tags/rl/ Recent content in RL on CheaSim Blog Hugo zh-cn Wed, 24 Jun 2026 21:30:00 +0000 https://www.cheasim.com/2026/06/24/skill0-arxiv-2604-02268/ Wed, 24 Jun 2026 21:30:00 +0000 https://www.cheasim.com/2026/06/24/skill0-arxiv-2604-02268/ <p>前两篇刚好写了 <a href="https://www.cheasim.com/2026/06/21/sga-mcts-arxiv-2604-14712/">SGA-MCTS</a> 和 <a href="https://www.cheasim.com/2026/06/22/skillx-arxiv-2604-04804/">SkillX</a>，今天这篇 <a href="https://arxiv.org/abs/2604.02268">SKILL0: In-Context Agentic Reinforcement Learning for Skill Internalization</a> 正好能把这条线补完整。</p> <p>如果说 SGA-MCTS 和 SkillX 都是在讨论“怎么把 Agent 经验放到外部系统里”，那 Skill0 问的是另一个更狠的问题：外部技能库能不能只在训练时用，最后把技能内化到模型参数里，让 Agent 测试时不再依赖 runtime skill retrieval？</p>