AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Rubric Curriculum RL: Exploiting the Generation-Verification Gap in Non-Verifiable Domains

Tejas Krishnan, Sumeet Ramesh Motwani, Charles London, Suhaas M Bhat, Huitian Jiao, Philip Torr, Riashat Islam, Christopher Summerfield, Christian Schroeder de Witt, Qilong Gu, Shital Shah

ICML 2026 · Reinforcement learning › RL from feedback · approach: Scaling RL with verifiable rewards

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 pairwise rewards instead of absolute rubric scores (ablation)RcRL · FictionBenchwin rate: 55.7 -> 73.9Improved
add curriculum over rubric criteria to pairwise RaR (ablation)Pairwise RaR · FictionBenchwin rate: 69.6 -> 73.9Improved
add rubric criteria curriculum to absolute RaRAbsolute RaR · FictionBenchwin rate: 57.5 -> 73.9Improved
add curriculum to absolute rewards (curriculum alone) (ablation)Absolute RaR · FictionBenchwin rate: 57.5 -> 55.7No clear effect
add pairwise rewards to curriculum on HealthBench (ablation)RcRL · HealthBench fullscore improvement: +13.40% -> +14.60%Improved
add pairwise rewards to curriculum on HealthBench consensus (ablation)RcRL · HealthBench consensusscore improvement: +23.87% -> +25.38%Improved
use RcRL with fixed rubric instead of instance-specific rubricsRcRL · HealthBench fullscore improvement: +13.52% -> +14.60%Improved
add weighting of newest criterion (ablation)RcRL · FictionBenchwin rate: 72.1 -> 73.9No clear effect
add KL term to RcRL (ablation)RcRL · FictionBenchwin rate: 73.9 -> 50.5Hurt
add exploration bonus to RcRL (ablation)RcRL · FictionBenchwin rate: 73.9 -> 60.0Hurt
vary curriculum ordering (ablation)RcRL · FictionBenchwin rate: 73.9; 79.6; 81.0Mixed
remove curriculum (RaR) vs RcRL on policy entropyRaR · LechMazur writing benchmarkentropy ratio: 0.73 -> 0.93Improved

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