←Home

    Models That Watch, Remember and Speak Up

    How vision-language models are learning to follow long and live video streams: they remember what they have seen, decide for themselves when to respond, and hold real-time audio-visual conversations.

    In order

    1 stories

    Photo: Umut Tuluoglu

    Latent · Video AI evaluation

    Over half the questions in 14 video AI benchmarks can be answered without real viewing

    Researchers at Sejong University and NAVER Cloud pared 24,416 video test questions to 11,033 that survived shortcut checks; on those, even the top-scoring model got fewer than half right.

    arXiv preprint · 2 July 2026