←back to thread

Show HN: A personal YouTube frontend based on yt-dlp

(github.com)
319 points modmodmod | 1 comments | 15 Mar 25 15:45 UTC | HN request time: 0.228s | source
Show context
greggyb ◴[15 Mar 25 18:17 UTC] No.43374252[source]▶
A question for the author or anyone else who has experience in similar solutions.

Is there any good solution for discovering new content? Much of the time, I want to stick to my subscriptions, but I do enjoy content surfaced by the algorithm at least once weekly, sometimes more often. My concern in taking my viewing off-platform is twofold: 1) going to YouTube will prompt me with all the stuff I've already watched off platform, and 2) any changes to my viewing habits won't be reflected in algorithmic suggestions.

Am I making any bad assumptions or missing anything that would be useful?

As an example, I usually get conference presentations surfaced for me, but I don't track conferences to know when I should go looking for presentations. YouTube is good at surfacing these for me.

replies(10): >>43374303 #>>43374312 #>>43374323 #>>43374347 #>>43374511 #>>43374543 #>>43374665 #>>43374717 #>>43374767 #>>43378703 #
toomuchtodo ◴[15 Mar 25 19:10 UTC] No.43374543[source]▶
If you were to have something local build you an algorithm, what signal would you want it to consume and how far from the median would you want it to deviate? Would you want it to use signal from online socials?
replies(2): >>43374781 #>>43375100 #
charcircuit ◴[15 Mar 25 20:56 UTC] No.43375100[source]▶
Why limit it to local? You could use the API for the YouTube recommendations. You already are using the YouTube API for the videos themselves.
replies(1): >>43375155 #
1. toomuchtodo ◴[15 Mar 25 21:06 UTC] No.43375155[source]▶
Certainly, ingest all the signal you’d like, and then emit a feed for clients to consume (or to be republished). Could run locally, could run in a container, could run on an AT protocol PDS. It is an algorithm/discovery/recommendation sovereignty play.