Tsinghua University

DeepSeek's Visual Primitives: Multimodal Reasoning From Seeing to Pointing

DeepSeek, PKU & Tsinghua released a framework making AI point at images while reasoning, then deleted the repo. It highlights the academia-product gap