AI RESEARCH
Not All Tokens Learn Alike: Attention Entropy Reveals Heterogeneous Signals in RL Reasoning
arXiv CS.CL
•
ArXi:2605.07660v1 Announce Type: new Reinforcement-learning-based post-