Beyond 80/20: High-Entropy Minority Tokens Drive Effective RL for LLM Reasoningarxiv.org3 points·mdp2021··0 commentsOpen articleSaveView on HN