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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| decouple an epistemic state from the interaction history (answer only from Et) (ablation) | SCOUT · LooGLE-v2 | accuracy (%): 70.5 -> 78.2 | Improved |
| add foraging actions that access the document (ablation) | SCOUT · LooGLE-v2 | accuracy (%): 17.7 -> 78.2 | Improved |
| add file tools beyond basic navigation (ablation) | SCOUT · LooGLE-v2 | accuracy (%): 74.2 -> 78.2 | Improved |
| add provenance grounding to state updates (ablation) | SCOUT · LooGLE-v2 | accuracy (%): 75.5 -> 78.2 | Improved |
| remove decoupled state (ReAct-style) (ablation) | SCOUT · LooGLE-v2 | token cost (k): 29.7 -> 28.3 | No clear effect |
| SFT of single-pass LLM on long context | Qwen2.5-72B-Instruct · LooGLE-v2 and InfiniteBench | accuracy (%): 24.0 -> 25.1; 49.4 -> 48.7 | No clear effect |
| SFT within SCOUT paradigm | Qwen2.5-72B-Instruct · LooGLE-v2 | accuracy (%): 28.3 -> 55.4 | Improved |
| SFT within SCOUT paradigm | Qwen2.5-72B-Instruct · InfiniteBench | accuracy (%): 53.2 -> 78.3 | Improved |
| SFT within SCOUT paradigm | Qwen3-30B-A3B-Instruct · LooGLE-v2 | accuracy (%): 41.2 -> 54.2 | Improved |
| SFT within SCOUT paradigm | Qwen3-30B-A3B-Instruct · InfiniteBench | accuracy (%): 65.4 -> 75.8 | Improved |
| DAPO in the SCOUT setting | Qwen2.5-72B-Instruct · LooGLE-v2 | accuracy (%): 28.3 -> 35.1 | Improved |
| DAPO in the SCOUT setting | Qwen3-30B-A3B-Instruct · LooGLE-v2 | accuracy (%): 41.2 -> 45.1 | Improved |
Closest papers by meaning
- Resolving Evidence Sparsity: Agentic Context Engineering for Long-Document Understanding (CVPR 2026 · similarity 0.88)
- Holistic Reasoning with Long-Context LMs: A Benchmark for Database Operations on Massive Textual Data (ICLR 2025 · similarity 0.87)
- GOLD PANNING: Strategic Context Shuffling for Needle-in-Haystack Reasoning (NeurIPS 2026 · similarity 0.87)
- Iterative Self-Incentivization Empowers Large Language Models as Agentic Searchers (NeurIPS 2025 · similarity 0.87)
- Scent of Knowledge: Optimizing Search-Enhanced Reasoning with Information Foraging (NeurIPS 2025 · similarity 0.87)
- Retriever-Free Retrieval-Augmented Reasoning via Corpus-Traversing MCTS (NeurIPS 2026 · similarity 0.86)
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.
