arXiv Artificial Intelligence· Kunlun Zhu, Tao Feng, Xu Han, Heng Wang, Ling Yang, Xiangru Tang, Jiaxuan You, Heng Ji·· 2 小时前AI 评分62
JevAgentHarness:让决策模型与 LLM 共享控制并训练 AgentDecision-4B
JevAgentHarness: Benchmarking, Deploying and Training Decision Models That Share Control with LLMs
AI 导读
研究提出 JevAgentHarness,把 Agent 每步的选项决策交给决策模型、由 LLM 在置信阈值以下兜底,并配套 AgentDecisionBench 与 DecisionGym 评测。
来源:arXiv Artificial Intelligence · arxiv.org