The authors assembled LongVideo, comprising more than 1,000 standalone long videos and two multi-season television series, totaling approximately 900 hours.
External memory tools improve models’ reasoning across long video streams
Jia and colleagues introduce a roughly 900-hour video benchmark and report performance gains above 50% for all 13 tested models when paired with their memory-and-retrieval system.
Chinese Tech
Yanhao Jia · Yi Zhu · Yiran Zhong · Weigao Sun · Feida Zhu · Pengkun Jiao · +4 more
Nanyang Technological University · Token Foundry, Alibaba · The Chinese University of Hong Kong · Institute of Neuroscience, Chinese Academy of Sciences · University of Chinese Academy of Sciences
Research Digest··2 min read
Jia and colleagues test whether AI models can connect evidence scattered across long videos, including multiple television seasons.
The Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.