Can LLMs Really Judge with Reasoning? Microsoft and Tsinghua Researchers Introduce Reward Reasoning Models to Dynamically Scale Test-Time Compute for Better Alignment

Reinforcement learning (RL) has emerged as a fundamental approach in LLM post-training, utilizing supervision signals from human feedback (RLHF) or verifiable rewards (RLVR). While RLVR shows promise in mathematical reasoning, it…