AI Paper Atlas

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

What LLMs Explain Is Not What They Believe: Evaluating Explanation Sufficiency Under Models' Own Input Beliefs

Nhi Nguyen, Shauli Ravfogel, Rajesh Ranganath

ICML 2026 · Large language models › Language, safety and interpretability · approach: LLM unlearning and benchmarks · 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
add few-shot examples with contradiction to the alteration prompt (ablation)SCSuff · IMDB (Qwen and Llama)SCSuff: drops near zeroHurt
add natural few-shot IMDB examples to the alteration prompt (ablation)SCSuff · IMDBSCSuffNo clear effect
add few-shot examples with contradiction to the alteration prompt (ablation)SCSuff · IMDB (Ministral)SCSuff: decreases substantially, not to zeroHurt
increase number of alternative inputs (ablation)SCSuff · MMLU + authority and IMDB, 8B modelsSCSuff: max diff 0.038No clear effect
paraphrase the alteration prompt (ablation)SCSuff · MMLU + authority and IMDB, 8B modelsSCSuff: max diff 0.057No clear effect
remove chain-of-thought from predictive model (ablation)SCSuff · MMLU + authority and IMDBSCSuff: max diff 0.074No clear effect
use a different LLM to generate alternative inputs (ablation)SCSuff · MMLU + authority and IMDBSCSuff: max diff 0.175Hurt
use final-token hidden state instead of average hidden state to predict SCSuff (ablation)probe · MMLU + authority and IMDB, 8B modelsprobe accuracy: accuracy < 0.8 -> above 0.84Improved
increase model sizeSCSuff · 9 LLMs, 4 datasetsSpearman correlation with SCSuff: rho = -0.23No clear effect
improve task accuracySCSuff · 9 LLMs, 4 datasetsSpearman correlation with SCSuff: rho = 0.07No clear effect
higher output entropySCSuff · 9 LLMs, 4 datasetsSpearman correlation with SCSuff: rho = 0.24No 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