田渊栋和Sergey Levine参与开发新型RL算法，能通过多轮训练让智能体学会协作推理
"ColBench" 相关文章
Meta 推出强化学习新框架 SWEET-RL，让 AI 更懂人类意图
Meta AI Researchers Introduced SWEET-RL and CollaborativeAgentBench: A Step-Wise Reinforcement Learning Framework to Train Multi-Turn Language Agents for Realistic Human-AI Collaboration Tasks
