AEWM 世界模型：让 LLM 智能体编辑任务状态而非模拟工具响应
Agent-Editing World Model: Rethinking World Modeling for LLM Agents
一篇很新颖的智能体论文：不模拟工具输出，而是让模型直接修改任务状态，Action Judge 拿了 70.5% F1，做 Agent 的值得看看思路。
论文提出 Agent-Editing World Model（AEWM），不再预测环境观察结果，而是建模推理和动作如何影响任务进度。AEWM 用 Action Judge 将决策区分为 Critical、Exploratory 和 Noisy 三类，在自建基准上取得 70.5% macro-F1，超过最强前沿基线 10.6 分。其 EditAct 模块直接修改影响后续决策的任务状态，在 6 个基准、3 个智能体骨干上平均提升 3.2-6.7 分。基于验证轨迹的拒绝采样微调版本 AEWM-RFT 在三个领域比 Self-RFT 再提升 2.2-2.6 分。训练覆盖 Search、Terminal 和软件工程三类场景。
Agent-Editing World Model: Rethinking World Modeling for LLM Agents
Recent advances in large language models (LLMs) have enabled agents to tackle long-horizon tasks across diverse environments. To further improve agent performance, existing language world models typically predict environment observations, yet reconstructing high-entropy, execution-dependent tool responses offers limited value when real feedback is available. Meanwhile, agents suffer from \emph{task-state contamination}, where unsupported assumptions and outdated plans persist in history and distort subsequent decisions. We propose the \textbf{Agent-Editing World Model (AEWM)}, which models how reasoning and actions shape future task progress rather than simulating tool responses. AEWM combines \textbf{Action Judge} to distinguish \textsc{Critical}, \textsc{Exploratory}, and \textsc{Noisy} decisions with \textbf{State Revision} to edit noisy reasoning--action continuations from the same observed history. \textbf{EditAct} integrates these capabilities with real execution, directly changing the state underlying subsequent decisions rather than merely providing critiques. We train AEWM across Search, Terminal, and Software Engineering through mid-training and supervised fine-tuning. AEWM achieves 70.5\% macro-F1 on our Action Judge benchmark, exceeding the strongest frontier baseline by 10.6 points. Across six benchmarks and three agent backbones, EditAct improves average scores by 3.2--6.7 points over the strongest baseline. Furthermore, rejection sampling fine-tuning on verified EditAct trajectories, termed \textbf{AEWM-RFT}, improves over Self-RFT by 2.2--2.6 points across three domains without online AEWM guidance.
