AI RESEARCH

PRISM: Pre-alignment via Black-box On-policy Distillation for Multimodal Reinforcement Learning

arXiv CS.AI

ArXi:2604.28123v1 Announce Type: cross The standard post-