Meta新论文：训练智能体学会何时使用记忆与工具，成功率翻倍
New Meta paper says giving an agent more memory and tools is not enough if the a…
做 Agent 的同学必看，这篇给出了训练智能体自主决定何时使用记忆与工具的具体方法，ALFWorld 上成功率从 47.9% 提到 96.9%，值得照着复现。
New Meta paper says giving an agent more memory and tools is not enough if the agent never learns when they are worth using.
Meta新论文指出，如果智能体从不学习何时使用记忆和工具，那么仅仅赋予其更多记忆和工具是不够的。
Long-running agents work better when they are trained to decide when external state is worth using.
当智能体被训练来决定何时使用外部状态时，长期运行的智能体表现更佳。
EvoHarness-RL took Qwen3-8B to 96.9% success on ALFWorld while harness use fell to about 1 call per episode.
EvoHarness-RL将Qwen3-8B在ALFWorld上的成功率提升至96.9%，而工具调用次数降至每集约1次。
Long tasks make agents forget current state, completed steps, and useful past experience. EvoHarness-RL keeps those outside the model as Belief, Progress, and Experience.
长期任务使智能体忘记当前状态、已完成步骤和有用的过往经验。EvoHarness-RL将这些信息作为信念、进度和经验保存在模型外部。
The policy learns when to read or update that state because every harness call consumes an interaction step.
策略学会何时读取或更新该状态，因为每次工具调用都消耗一个交互步骤。
Qwen3-8B with ReAct scored 47.9% on the seen ALFWorld split. EvoHarness-RL reached 96.9%; on unseen tasks, it reached 86.6% versus 50.0% for ReAct.
使用ReAct的Qwen3-8B在已见ALFWorld分割上得分为47.9%。EvoHarness-RL达到96.9%；在未见任务上，其得分为86.6%，而ReAct为50.0%。
During RL training, frequent state and memory calls gradually fell to roughly 1 per episode while success kept rising. Routine behavior appears to move into the model, leaving external state for moments when it is useful.
在强化学习训练期间，频繁的状态和记忆调用逐渐降至每集约1次，而成功率持续上升。常规行为似乎内化到模型中，外部状态仅在有用时被调用。
Agent builders should train the policy for when to consult and update memory instead of hard-coding constant access.
智能体构建者应训练策略决定何时查阅和更新记忆，而非硬编码持续访问。
– arxiv. org/abs/2608.05446
– arxiv.org/abs/2608.05446
Title: "EvoHarness-RL: Learning Self-Evolving Runtime Harness for Long-Horizon LLM Agents"
标题：“EvoHarness-RL：学习自进化运行时工具以支持长周期LLM智能体”
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力