AI Paper Atlas

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

The Truth Stays in the Family: Enhancing Contextual Truthfulness via Inherited Heads in Model Lineages

Miso Choi, Seonga Choi, Mincheol Kwon, Woosung Joung, Jinkyu Kim, Jungbeom Lee

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Efficient LLM fine-tuning systems

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
apply Truth Score soft head gating (TruthProbe) to the same LLMVicuna-7B · HaluEvalF1: 13.37 -> 29.15Improved
apply base-LLM Truth Scores as gate to finetuned Vicuna-7BVicuna-7B · HaluEvalF1: 13.37 -> 57.17Improved
transfer base-LLM Truth Scores to Qwen2.5-7B-InstructQwen2.5-7B-Instruct · HaluEvalF1: 16.29 -> 17.24Improved
transfer base-LLM Truth Scores to LLaVA-1.5LLaVA-1.5 · POPE (COCO)accuracy: 86.9 -> 86.7No clear effect
transfer base-LLM Truth Scores to Qwen2.5-VL-OmniQwen2.5-VL-Omni · POPE (COCO)accuracy: 85.1 -> 87.3Improved
transfer base-LLM Truth Scores to LLaVA-1.5LLaVA-1.5 · CHAIRCHAIRS: 23.00 -> 17.40Improved
transfer base-LLM Truth Scores to Qwen2.5-VL-OmniQwen2.5-VL-Omni · CHAIRCHAIRI: 5.26 -> 5.94Hurt
derive Truth Scores from the MLLM itself instead of its base LLMLLaVA-1.5 · CHAIRCHAIRS: 17.40 (base LLM) vs 21.60 (MLLM)No clear effect
replace Truth Score gate with random head gating (ablation)LLaVA-1.5 · POPE (COCO)accuracy: 86.9 -> 86.1Hurt
replace Truth Score gate with random head gating (ablation)LLaVA-NeXT · POPE (COCO)accuracy: 87.7 -> 87.1Hurt
compare TruthProbe to ITI on parametric-knowledge taskLLaMA2-7B-Chat · TruthfulQA generationtruthfulness: 56.40 -> 56.91 (ITI 57.64)No clear effect
TruthProbeFT LLM on TruthfulQALLaMA2-7B-Chat · TruthfulQA generationinformativeness: 25.56 -> 29.02Improved

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