AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

Information Flow Reveals When to Trust Language Models

Rui Xu, Yi Chen, Jiujiu Chen, Sihong Xie

ICML 2026 · Large language models › Retrieval and knowledge · approach: Retrieval-augmented reasoning

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
model progressive layer-wise information flow (simulatability and concentration) instead of single-layer or pooled attention signalsRAG uncertainty quantification · SQuAD2.0, HotpotQA, MS MARCOAUROC: average AUROC 0.709 vs runner-up 0.676Improved
use the information-flow method instead of Focus on HotpotQALLaMA-3.2-3B-Instruct · HotpotQA testAUROC: 0.701 (Focus) vs 0.672 (Ours)Hurt
train the calibrator on a different dataset than the test set (ablation)Information-flow calibrator · Cross-dataset (SQuAD2.0, HotpotQA, MS MARCO)AUROC: SQuAD2.0 test, LLaMA-3.2-3B: 0.735 / 0.728 / 0.715 across train setsNo clear effect
ablate the top-5 context tokens ranked by the flow measures (ablation)Information-flow rankings · 100 correct samples per datasetremaining correct predictions: over 50% of correct predictions become incorrectHurt
ablate the bottom-5 context tokens ranked by the flow measures (ablation)Information-flow rankings · 100 correct samples per datasetremaining correct predictionsNo clear effect
human-verify ranker relevance layouts before computing simulatability (ablation)Simulatability · Ranker bias checkAUROC and AUPRC: modest increaseImproved
apply the same post-hoc calibration to baseline methods (ablation)Baseline UQ methods · Appendix KAUROCNo clear effect

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