Have you seen DeepMark... https://news.ycombinator.com/item?id=49288896 ?
Looks well done - people may bookmark vids and want to use all their YT bookmarks as a source for the sauce. Get all the reading done in one shot.
Lastly, could a bookmarklet prompt an AI to summarize a video by pre-filling a prompt? I suppose that's what is being done here, but via hover windows.