1 paper
Reuben Tan, Baolin Peng, Zhengyuan Yang +16
Agentic reasoning models trained with multimodal reinforcement learning (MMRL) have become increasingly capable, yet they are almost universally optimized using sparse, outcome-bas…