AI Paper Atlas

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

Harnessing Reasoning Trajectories for Hallucination Detection via Answer-agreement Representation Shaping

Jianxiong Zhang, Bing Guo, Yuming Jiang, Haobo Wang, Bo An, Xuefeng Du

ICML 2026 · Large language models › Language, safety and interpretability · approach: Reasoning mechanisms and failure modes · 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
shape embeddings with ARS instead of vanilla LRM embeddings (CCS)CCS detector · TruthfulQA, Qwen3-8BAUROC: 66.85 -> 86.64Improved
shape embeddings with ARS (CCS)CCS detector · TriviaQA, Qwen3-8BAUROC: 59.24 -> 88.54Improved
shape embeddings with ARS (HaloScope)HaloScope · TruthfulQA, Qwen3-8BAUROC: 57.78 -> 71.03Improved
shape embeddings with ARS (EigenScore)EigenScore · TriviaQA, Qwen3-8BAUROC: 55.26 -> 56.09Improved
use latent trace-boundary intervention instead of token deletion (ablation)ARS (CCS) · TruthfulQA, Qwen3-8BAUROC: 75.25 -> 86.64Improved
use latent intervention instead of trace paraphrasing (ablation)ARS (CCS) · TruthfulQA, Qwen3-8BAUROC: 50.47 -> 86.64Improved
use learned ARS shaping instead of PCA projection (ablation)ARS (CCS) · TruthfulQA, Qwen3-8BAUROC: 73.26 -> 86.64Improved
use learned ARS shaping instead of random projection (ablation)ARS (CCS) · TruthfulQA, Qwen3-8BAUROC: 68.36 -> 86.64Improved
perturb only the last token instead of last 1% tokens (ablation)ARS (CCS) · TruthfulQA, Qwen3-8BAUROC: 81.70 -> 86.64Improved
use greedy instead of multinomial sampling (ablation)ARS (CCS) · TruthfulQA, Qwen3-8BAUROC: 84.49 -> 86.64Improved
use embeddings f instead of Attn(f) (ablation)ARS (CCS) · MATH-500, Qwen3-8BAUROC: 67.79 -> 78.66Improved
increase shaping temperature from 0.1 to 0.5 (ablation)ARS (CCS) · TruthfulQA, Qwen3-8BAUROC: 86.64 -> 63.97Hurt

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