Building a Precise Video Language with Human-AI Oversight

arXiv:2604.21718v1 Announce Type: cross Abstract: Video-language models (VLMs) learn to reason about the dynamic visual world through natural language. We introduce a suite of open datasets, benchmarks, and recipes for scalable oversight that enable precise video captioning. First, we define a...