PANDO: Efficient Multimodal AI Agents via Online Skill Distillation 事件

PANDO: Efficient Multimodal AI Agents via Online Skill Distillation arXiv:2605.24785v1 Announce Type: new Abstract: Recent advances in multimodal web agents often rely on increased inference-time computation, including rollout search, verifier passes, offline skill discovery, and specialist model stacks. This raises a central question: can a web agent become more efficient as it accumulates experience, rather than more expensive? We first analyze trajectories from VisualWebArena and identify th