Earlier quoted context omitted.
My half-serious and half-assed attempt at solving a similar problem: make a bot watch the video for you first, then skip the crap. https://butter.sonnet.io
That is such a well made website, love the art. So it can tell when videos contain sponsored segments just from the video's content? What parameters does it check to make its conclusion?
It's quite primitive:
1. I have a service extracting captions from YT videos.
2. I process them a little bit (parallel batches with a sliding window containing the captions),
3. then add to a simple GPT-4 prompt with some examples.
Feel free to check the repo source (esp. detector.js and its dependencies)