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

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.

good question. I don't think I have a definitive answer but I'll try:

- pure luck. sometimes I discover a channel/creator/blog by pure accident, I'm an avid rss reader and HN adept so content comes to me naturally, so to speak.

- following a feed (be it a website's rss feed, reddit/YouTube) sometimes made me discover related feeds, simply because someone wrote about a cool project a peer made and links their YouTube/github/blog

I view Discovery as a social problem where the content you want is almost always clustered between a relatively small number of creators, regions, etc.

Technically it then becomes less of an indexing everything problem and more of a find a few cornerstone creators, say Khan academy, and occasionally branching out.

So to answer your question I don’t thing the cost/benefit for automating discovery is much better then spending 20 minutes and finding enough cornerstones to fill you for 100+ hours of content. Or similarly finding a social group like an rss feed, say in ios development it would be fatbobman, and sourcing it from there.

Time to source content isn’t the bottleneck worthy of software solutions, yet for monetization reasons discovery is the vice grip of social media and made to be the most important thing.

I looked into this as well since I find the YouTube algorithm terrible, but couldn’t find any API for exploration. Which makes sense they want to control what you watch and hence monetize. In a perfect world you could just pick an open source recommendation algorithm from a marketplace and YouTube would just be a wrapper around s3 buckets and some index.

I've been using a third party app to watch the videos and the official app to discover content.

Instead of just clicking the video I click share and watch on the unofficial with no ads.

There’s a lot of truth to this but one of the most powerful elements of a discovery algorithm is finding things you completely did not look for, ie Christopher Columbus and the western continents. Like your cornerstones are iOS and recipe videos but you discover the right dance video and it changes your whole life.

I've been using Unhook[0] for years that it's almost a jumpscare for me to see a recommended video or the Youtube homepage. Your social circles and natural serendipity should be plenty for finding new creators. And in general, avoiding algorithmic feeds will help with ADHD and mindless scrolling.

[0] https://unhook.app/

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?

An even more perfect world would not have S3 buckets.

You have to store bits somewhere, and an S3 compatible target optimizes for flexibility and optionality. It can be local (Minio), it can be remote, the client does not care where it is. Even the Internet Archive's API is S3-ish.

Check out the Vinegar extension if you use Safari. Same old YouTube but all the videos are replaced with HTML5 <video>s.

I use a Firefox profile to watch specific videos while logged-out just for the focused recommendations.

I've also noticed that I getting more recommendations for small creators with little to no views/subs when I'm browsing from a smaller, developing country.

I readily follow youtube links offered on HN discussions. If anything, I could use more of these.

But otherwise I agree with your concern. Video recommendations on youtube was far from perfect (very repetitive in my experience), but was uncovering useful stuff.

I owe many interests in my life to the little recommendation tab next to a currently playing video on youtube

This is a good idea. One signal would be HN mentions. Second might be reddit mentions, but with a lot of qualifications.

As a first step, a page showing recent youtube links from HN would be nice!

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.

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.

> you discover the right dance video and it changes your whole life

You're going to have to explain this one, how would a dance video change my life? Being exposed to something new that becomes profoundly life changing seems like a romanticized notion and not a realistic one especially within a monetized environment.

We're exposed to new stuff everyday, just because .0001% is truly impactful doesn't justify watching 100_000 short reels of ads, even if Google and Facebook REALLY want us to.

I.e., that is (remember is and i)

E.g., for example (remember example and e)

Well I’m sure there others who will agree that something small and completely unexpected has had a profound influence in their lives. The simplest example is something so novel and interesting opens you into a deep rabbit hole that changes your career and or who you meet, befriend or marry. The lack of a good recommendation algorithm is exactly the problem where these content platforms is you feel like you have to watch 100k videos to have a chance at such an encounter.

I am almost a month into having a Perplexity subscription and I am not sure I can not have a deep research subscription at this point.

I have found youtube videos this month that I don't know how I would have found otherwise that were just part of the sources for what deep research came back with.

It has really created the opposite problem for me is I have so much good information I don't even know what to do with it right now. I am probably taking a month off to just sort through what I found this past month.

Does this have an apparent impact on your recommendations?

I believe I may have found something that is headed this way:

https://www.graze.social/

https://www.graze.social/blog/grazer-algorithm-engine

https://github.com/graze-social/grazer