AI Paper Atlas

Research map / Large language models / Alignment and preferences

Robust Reward Modeling via Causal Rubrics

Pragya Srivastava, Harman Singh, Rahul Madhavan, Gandharv Patil, Sravanti Addepalli, Arun Suggala, Rengarajan Aravamudhan, Soumya Sharma, Anirban Laha, Aravindan Raghuveer, Karthikeyan Shanmugam, Doina Precup

ICLR 2026 · Large language models › Alignment and preferences · approach: Reward model design and preference models · 13 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
train with causal and neutral augmentations (Crome) instead of RRMreward model training · RewardBench PairPM (Gemma-2-9B)average accuracy: 82.54 -> 87.84Improved
train with causal and neutral augmentations (Crome) instead of RRMreward model training · RewardBench PairPM (Gemma-2-9B) Safetyaccuracy: 74.70 -> 87.14Improved
train with Crome instead of RRMreward model training · RewardBench PairPM (Gemma-2-2B)average accuracy: 66.23 -> 70.69Improved
train with Crome instead of RRMreward model training · RewardBench PairPM Chat (Qwen2.5-7B)accuracy: 97.21 -> 96.37Hurt
train with Crome instead of RRMreward model training · RewardBench BT Chat-Hard (Gemma-2-9B)accuracy: 69.15 -> 65.83Hurt
train with Crome instead of RRMreward model training · reWordBench PairPM (Gemma-2-9B)average accuracy: 64.68 -> 72.71Improved
train with Crome instead of RRMreward model training · reWordBench BT (Gemma-2-9B)average accuracy: 65.69 -> 69.81Improved
train with Crome instead of RRMreward model training · RewardBench to reWordBench accuracy drop (PairPM)percentage drop: 21.54% -> 19.78%Improved
use Crome reward model for Best-of-N selection instead of RMBest-of-N selection · WildGuardTest (N=32)attack success rate: 42.11 -> 39.39Improved
use Crome reward model for Best-of-N instead of RMBest-of-N selection · RewardBench BoN win rate (N=32)win rate vs RM: 44.79 vs 21.88Improved
use Gemma-3-27B instead of Gemini 2.0 Flash as oracle LLM for augmentationsCrome oracle choice · RewardBench PairPM (Gemma-2-9B)average accuracy: 82.53 -> 85.15Improved
add causal upgrade/degrade pairs plus irrelevant-query neutrals (IQN) to causal-only augmentation (ablation)Crome · RewardBench Chat-HardaccuracyImproved

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