Mengdi Wang

VideoGen-Agent is presented, a multimodal agent trained through multitask agentic reinforcement learning to use external tools for video generation to support learning tool use across video-generation tasks and show that the trained agent can benefit from subsequent advances in generation tools.

Skills stabilize agent execution by transforming noisy trajectories into procedural anchors, but they can fail under brittle assumptions and incompatible contexts.

LLMs align even better with human preferences when trained on *less* data, revealing that preference signals are surprisingly concentrated in the initial tokens of responses.