AI Paper Atlas

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

TRACEDET: HALLUCINATION DETECTION FROM THE DECODING TRACE OF DIFFUSION LARGE LANGUAGE MODELS

Shenxu Chang, Junchi Yu, Weixing Wang, Yongqiang Chen, Jialin Yu, Philip Torr, Jindong Gu

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Data and hallucination detection · 8 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
add sub-trace masking (IB-based sub-instance extraction) to the detector (ablation)TraceDet, LLaDA-8B-Instruct · Average of 3 QA datasetsAUROC (%): 69.1 -> 72.0Improved
add sub-trace masking (ablation)TraceDet, Dream-7B-Instruct · Average of 3 QA datasetsAUROC (%): 78.4 -> 80.8Improved
replace average stepwise entropy with the learned TraceDet detector (ablation)TraceDet, LLaDA-8B-Instruct · Average of 3 QA datasetsAUROC (%): 62.8 -> 72.0Improved
replace average stepwise entropy with the learned TraceDet detector (ablation)TraceDet, Dream-7B-Instruct · Average of 3 QA datasetsAUROC (%): 65.3 -> 80.8Improved
add masking on TriviaQA with 64 generation steps (ablation)TraceDet, Dream-7B-Instruct · TriviaQA (64)AUROC (%): 87.1 -> 86.7No clear effect
use TraceDet versus strongest baseline EigenScoreTraceDet, LLaDA-8B-Instruct · Average of 3 QA datasetsAUROC (%): 63.2 -> 72.0Improved
use TraceDet versus strongest baseline EigenScoreTraceDet, Dream-7B-Instruct · Average of 3 QA datasetsAUROC (%): 69.8 -> 80.8Improved
use TraceDet versus multi-sampling baselines for detection costTraceDet, LLaDA-8B-Instruct · Inference time, 100 samplestime (s): 801.35 -> 147.52Improved
increase generation length to 128 from moderate lengths (ablation)TraceDet, Dream-7B-Instruct · TriviaQAAUROCHurt
vary step length over 1, 2, 4, 8 (ablation)TraceDet, Dream-7B-Instruct · TriviaQA/HotpotQA/CommonsenseQAAUROCNo clear effect
use random remasking instead of low-confidence/entropy/top-k (ablation)TraceDet, Dream-7B-Instruct · TriviaQAAUROCHurt
vary remasking among low-confidence, entropy and top-k (ablation)TraceDet, Dream-7B-Instruct · TriviaQA and CommonsenseQAAUROC: 75-85% AUROCNo 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