AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Fine-Tuned LLMs Know They Don’t Know: A Parameter-Efficient Approach to Recovering Honesty

Zeyu Shi, Ziming Wang, Tianyu Chen, Shiqi Gao, Haoyi Zhou, Qingyun Sun, Jianxin Li

AAAI 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning generalization and delta compression · 2 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 Hessian-guided compensation (Stage 2) (ablation)HCNR · avg F1, FT on HotpotQAF1: 65.96 -> 72.84Improved
add Hessian-guided compensation (Stage 2) (ablation)HCNR · avg F1, FT on MedMCQAF1: 65.68 -> 73.02Improved
add Hessian-guided compensation (Stage 2) (ablation)HCNR · avg RF delta, FT on HotpotQARefusal delta: +33.09 -> +42.64Improved
use honesty-priority neuron selection instead of random (Stage 1) (ablation)HCNR · avg F1, FT on HotpotQAF1: 65.44 -> 72.84Improved
use honesty-priority neuron selection instead of random (Stage 1) (ablation)HCNR · avg F1, FT on MedMCQAF1: 67.39 -> 73.02Improved
add task-importance term to neuron priority (ablation)HCNR · domain accuracy, FT on HotpotQAdomain accuracy: 28.30 -> 30.30Improved
add task-importance term to neuron priority (ablation)HCNR · avg F1, FT on HotpotQAF1: 70.43 -> 72.84Improved
add task-importance term to neuron priority (ablation)HCNR · domain accuracy, FT on MedMCQAdomain accuracy: 9.32 -> 10.33Improved
use both random selection and no compensation (ablation)HCNR · avg F1, FT on HotpotQAF1: 72.84 -> 54.21Hurt
restore honesty with HCNR instead of RAITLlama-3.1-8B-Instruct FT on HotpotQA · domain accuracyaccuracy: 27.05 -> 30.30Improved
use HCNR instead of DPO for recovery timeLlama-3.1-8B-Instruct FT on HotpotQA · timeminutes: 42.78 min -> 3.93 minImproved
use HCNR (256 samples) instead of RAIT (5000 samples)Llama-3.1-8B-Instruct FT on HotpotQA · avg F1F1: 70.58 -> 72.84Improved

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