Alternating Reinforcement Learning for Rubric-Based Reward Modeling in Non-Verifiable LLM Post-Training

OpenRubrics: Towards Scalable Synthetic Rubric Generation for Reward Modeling and LLM Alignment