AEWM: editing agent task states beats simulating tool responses
RUC · hf · 2026-09-25
Existing language world models predict environment observations, but reconstructing high-entropy, execution-dependent tool responses adds little value when real feedback exists — and agents suffer from task-state contamination, where unsupported assumptions and outdated plans distort later decisions.
RUC's Agent-Editing World Model (AEWM) instead models how reasoning and actions shape future task progress. Key components: an Action Judge distinguishing Critical/Exploratory/Noisy decisions, State Revision to edit noisy reasoning-action continuations, and EditAct which integrates these with real execution to directly change the state underlying subsequent decisions.
Trained across Search, Terminal, and Software Engineering: 70.5% macro-F1 on their Action Judge benchmark (+10.6 over the strongest frontier baseline); EditAct improves average scores by 3.2–6.7 points across six benchmarks and three agent backbones; rejection-sampling fine-tuning (AEWM-RFT) adds another 2.2–2.6 points over Self-RFT.
More from coding & agent
- Meta launches Muse Code terminal coding agent, resets usage limits for DevDay — armand_ruiz · 2026-09-25
- User ditches GUI: local LLM now his main PC interface for installs, GRUB and more — Forward_Jackfruit813 · 2026-09-25
- Where an agent loop runs, how it fails, and how to build durable sessions — tempNull · 2026-09-25
- Community turns the Opus 5.5 pixel-art prompt into an open-source pixel-anims skill — majidmanzarpour · 2026-09-25
- Opus 5.5 + one prompt generates animated pixel art wizard purely in code — majidmanzarpour · 2026-09-25
- A visual harness for Claude Code: watching your agents work after hours — EricBuess · 2026-09-25