AI Paper Atlas

Research map / Large language models / Alignment and preferences

CAMEL: Confidence-Gated Reflection for Reward Modeling

Zirui Zhu, Hailun Xu, Yang Luo, Yong Liu, Kanchan Sarkar, Kun Xu, Yang You

ICML 2026 · Large language models › Alignment and preferences · approach: Reward model design and preference models · 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 reflection to all test inputsCAMEL-Fast · RewardBenchaccuracy: 90.5 -> 92.8Improved
add reflection to all test inputsCAMEL-Fast · RM-Benchaccuracy: 74.8 -> 84.2Improved
add reflection to all test inputsCAMEL-Fast · JudgeBenchaccuracy: 65.2 -> 71.6Improved
add reflection to all inputs (ablation)Qwen3-14B base · Avg of three benchmarksaccuracy: 71.9 -> 73.8Improved
add reflection to all inputs (ablation)Qwen3-14B-SFT · RewardBenchaccuracy: 90.6 -> 90.5No clear effect
add reflection to all inputs (ablation)Qwen3-14B-GRPO · RewardBenchaccuracy: 91.2 -> 90.0Hurt
add reflection to all inputs (ablation)Qwen3-14B-GRPO · JudgeBenchaccuracy: 62.9 -> 74.2Improved
add SFT on preference data (ablation)Qwen3-14B · RewardBenchaccuracy: 81.9 -> 90.6Improved
add counterfactual prefix augmentation and confidence gating to GRPO + reflection (ablation)CAMEL · Avg of three benchmarksaccuracy: 82.7 -> 81.1Hurt
add counterfactual prefix augmentation to GRPO (ablation)CAMEL · RewardBenchaccuracy: 91.2 -> 92.4Improved
raise confidence threshold tau from 3 to 7 (ablation)CAMEL · RM-Benchaccuracy: 81.3% -> 87.2%Improved
raise confidence threshold tau from 5 to 7 (ablation)CAMEL · JudgeBenchaccuracy: 67.1% -> 66.0%Hurt

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