AI Paper Atlas

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

Why Are LLMs Confidently Wrong? Correcting Overconfident Errors via Causal Head Intervention

Jing Ren, Bowen Li, Ziqi Xu, Xuechao Yang, Feng Xia

NeurIPS 2026 · Large language models › Language, safety and interpretability · approach: Uncertainty quantification for LLMs · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes CoCHI, a class-oriented causal head intervention method that targets attention heads causing overconfident LLM errors, rather than adjusting outputs post hoc or retraining.

Problem

LLMs often make overconfident errors, assigning high confidence to incorrect predictions, which undermines reliability; existing calibration methods act at the output level via post-hoc adjustment or retraining without addressing underlying causes.

Approach

Identifies internal attention heads causally responsible for overconfident errors and intervenes on them in a class-oriented manner, addressing miscalibration at its internal source rather than at the output level.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
apply CoCHI head masking to the modelLLaMA-3.1-8B · MMLUECE: 0.0569 (TS) -> 0.0490 (CoCHI)Improved
apply CoCHI head masking to the modelMistral-7B · MedMCQAECE: 0.0604 (TS) -> 0.0383 (CoCHI)Improved
apply CoCHI head masking to the modelGemma-2-9B · MMLUaccuracy: 0.6653 (TS) -> 0.6730 (CoCHI)Improved
apply CoCHI instead of TS on LLaMA-3.1-8B and Mistral-7BLLaMA-3.1-8B · OpenBookQAaccuracy: 0.6827 (TS) -> 0.6700 (CoCHI)Hurt
use ITI instead of CoCHI as the calibration methodLLaMA-3.1-8B and Mistral-7B · MMLUECE: 0.0490 (CoCHI) vs 0.2014 (ITI) on LLaMA; 0.3693 ITI on MistralHurt
mask only wrong-high facilitating heads (ablation)CoCHI · MMLUaccuracy: 0.6120 -> 0.2044Hurt
use the broad canonical union mask without top-k selection (ablation)CoCHI · MMLUECE: 0.0490 -> 0.3850Hurt
add causal head selection instead of random size-matched masking (ablation)CoCHI · MMLUECE: 0.4330 -> 0.0490Improved
add wrong-high facilitating heads and top-k ranking to the correct-high interfering-only mask (ablation)CoCHI · MMLUaccuracy: 0.4267 -> 0.6120Improved
mask only correct-high non-facilitating heads (ablation)CoCHI · MMLUECE: 0.0490 -> 0.1518Hurt
raise confidence threshold tau_c from 0.5 to 0.7 or 0.9 (ablation)CoCHI · MMLUECE: 0.0490 -> 0.0705 (0.7), 0.0696 (0.9)Hurt
lower confidence threshold tau_c to 0.2 (ablation)CoCHI · MMLUaccuracy: 0.6120 -> 0.5933Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Experimental setup

Baselines: post-hoc calibration methods retraining-based calibration methods

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