AI Paper Atlas

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

Rewarding Doubt: A Reinforcement Learning Approach to Calibrated Confidence Expression of Large Language Models

David Bani-Harouni, Chantal Pellegrini, Paul Stangel, Ege Özsoy, Kamilia Zaripova, Nassir Navab, Matthias Keicher

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning for reasoning data

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
fine-tune verbalized confidence with log-scoring-rule PPO rewardLlama-3-8B-Instruct · TriviaQAECE: 0.3459 -> 0.0226Improved
fine-tune verbalized confidence with log-scoring-rule PPO rewardLlama-3-8B-Instruct · TriviaQAAUROC: 0.5858 -> 0.8592Improved
fine-tune confidence with RL rewardLlama-3-8B-Instruct · TriviaQAaccuracy: 0.6310 -> 0.6309No clear effect
use RL instead of supervised trained probeRewarding Doubt · TriviaQAECE: 0.0189 -> 0.0226 ECE; AUROC 0.8173 -> 0.8592Mixed
apply Rewarding Doubt to multi-answer trainingLlama-3-8B-Instruct · QAMPARIECE: 0.5319 -> 0.0816Improved
apply Rewarding Doubt to other base models (ablation)Qwen-2.5-3B, Qwen-2.5-7B, Gemma-2-9B, LLaMA-3.1-8B · TriviaQAECE: Qwen-2.5-3B 0.5330 -> 0.1483Improved
train on TriviaQA, evaluate out of domainRewarding Doubt · MedQAECE: 0.4480 -> 0.1145Improved
train on TriviaQA, evaluate out of domainRewarding Doubt · CommonsenseQAECE: 0.2820 -> 0.2930No clear effect
train on single-answer, evaluate on multi-answer QAMPARI (ablation)Rewarding Doubt · QAMPARI multi factECE: 0.5319 -> 0.1777Improved
train specifically on multi-fact data rather than single fact (ablation)Rewarding Doubt · QAMPARI multi factECE: 0.1777 -> 0.1061Improved

Later papers that compared with this method

1 later accepted paper reports RewardingDoubt as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

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.

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