RARE: Reinforcement Learning with Adaptive Rubric Evolution for Open-Ended Generation
Verifiable rewards have advanced reinforcement learning in mathematics and code.
ProofPaper ↗
Key points
- Training therefore needs feedback on requirement satisfaction across different valid responses.
- Rubric-based reinforcement learning provides this feedback by aggregating scores from explicit criteria.
- Dynamic methods revise shared rubrics or select evolving criteria to restore discrimination, but may also change useful criteria.
- To address this, we propose Reinforcement Learning with Adaptive Rubric Evolution (RARE).
Sources (1)
- [1]RARE: Reinforcement Learning with Adaptive Rubric Evolution for Open-Ended GenerationarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 18, 05:54 PM
Verifiable rewards have advanced reinforcement learning in mathematics and code.
Training therefore needs feedback on requirement satisfaction across different valid responses.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 17, 2026ProbePO: Sparse Probes, Redistributed Credit for Critic-Free LLM Reinforcement Learning
- Oct 17, 2026Axiomatic World Modeling for Physics Reasoning
- Oct 16, 2026Training-Free Prompt Selection for Efficient Reinforcement Learning With Partial Rollouts
- Oct 16, 2026Welcome Beam: a 501B open model you can own, customize, and scale
- Oct 16, 2026Tree-Aware Posterior Distillation for Long-Horizon Agentic Reinforcement Learning
- Oct 16, 2026Language Models Learn to Improve Language Models