A 4B open model beats GPT-5 at finding when things happen in video

Ask

Ask about this presentation

Answers are generated from this presentation.

Chapters

1. 1A 4B open model beats GPT-5 at finding when things happen in video0:00
2. 2It points at the right span more often than GPT-5 does0:12
3. 3Eleven wins, and the big ones are all the same kind0:34
4. 4It compresses the video before the language model sees it0:55
5. 5The encoder tax is paid up front1:41
6. 6Weights and all three training sets, Apache-2.01:55
7. 7Ask a video model when, not just what2:11