One paper accepted in KDD 2026. Jizhou Guo, Zhaomin Wu, Hanchen Yang, Philip S. Yu. Mining Intrinsic Rewards from LLM Hidden States for Efficient Best-of-N Sampling