AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

SCOUT: Active Information Foraging for Long-Text Understanding with Decoupled Epistemic States

Zhenliang Zhang, Wenqing Wang, Yong Hu, Yaming Yang, Jiaheng Gao, Chen Shen, Xiaojun Wan

ICML 2026 · Large language models › Language, safety and interpretability · approach: Faithfulness of LLM explanations · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
decouple an epistemic state from the interaction history (answer only from Et) (ablation)SCOUT · LooGLE-v2accuracy (%): 70.5 -> 78.2Improved
add foraging actions that access the document (ablation)SCOUT · LooGLE-v2accuracy (%): 17.7 -> 78.2Improved
add file tools beyond basic navigation (ablation)SCOUT · LooGLE-v2accuracy (%): 74.2 -> 78.2Improved
add provenance grounding to state updates (ablation)SCOUT · LooGLE-v2accuracy (%): 75.5 -> 78.2Improved
remove decoupled state (ReAct-style) (ablation)SCOUT · LooGLE-v2token cost (k): 29.7 -> 28.3No clear effect
SFT of single-pass LLM on long contextQwen2.5-72B-Instruct · LooGLE-v2 and InfiniteBenchaccuracy (%): 24.0 -> 25.1; 49.4 -> 48.7No clear effect
SFT within SCOUT paradigmQwen2.5-72B-Instruct · LooGLE-v2accuracy (%): 28.3 -> 55.4Improved
SFT within SCOUT paradigmQwen2.5-72B-Instruct · InfiniteBenchaccuracy (%): 53.2 -> 78.3Improved
SFT within SCOUT paradigmQwen3-30B-A3B-Instruct · LooGLE-v2accuracy (%): 41.2 -> 54.2Improved
SFT within SCOUT paradigmQwen3-30B-A3B-Instruct · InfiniteBenchaccuracy (%): 65.4 -> 75.8Improved
DAPO in the SCOUT settingQwen2.5-72B-Instruct · LooGLE-v2accuracy (%): 28.3 -> 35.1Improved
DAPO in the SCOUT settingQwen3-30B-A3B-Instruct · LooGLE-v2accuracy (%): 41.2 -> 45.1Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy