Own Your HackerTube: Build Your Own AI-Filtered Video Feed
What a recommendation feed pushes at you is not what you asked for. I built Own Your HackerTube: write your standards, follow a few YouTube channels, and an AI filters every day and learns your taste from your feedback.
In the AI era, you can change a lot of things you could not change before. What you watch every day is one of them.
Where it started: fed up with the feed
You open YouTube or Bilibili to learn something, scroll for an hour, and remember almost nothing. A recommendation feed wants you to stay a little longer. It does not necessarily want you to get something out of it.
Andrej Karpathy built HackerTube. As I read it, he was also tired of the video recommendations on the big platforms, so he hand-picked high-quality videos from Hacker News and turned them into a clean video site.
The idea
I borrowed his approach, and an idea from Liu Xiaopai, a well-known voice in the Chinese indie community: don't let AI summarize a video for you. Use AI to filter instead, so that only what you really care about reaches you. In his words (my translation): "Let AI cut down the time I spend searching, and add to the time I spend watching what I actually care about." His article is here, in Chinese.
So I built Own Your HackerTube.
How it works
- Tell the AI your standards. Plain words are enough. You can also say what you don't want, for example: "no clickbait titles, and nothing that only explains concepts without real cases or numbers."
- Tell it which channels you follow. Right now that means YouTube channels.
- The AI watches for you, every day. It checks the new videos from those channels against your standards. The ones that pass are published to your own YouTube-style site, with a reason why each is worth your time and the key moments from the creator's chapters that you can jump to.
This is not a summary. It decides what deserves your time first, and you still do the watching.
Your feedback makes it understand you
On any pick, you can give feedback later: a great pick, or not a fit.
If it is not a fit, say why: too basic, off topic, clickbait or only concepts, not a source you trust, or something you already know. The AI remembers your judgment and adds it to the criteria for future picks.
Over time this becomes your own source of quality input, with your taste and preferences built in.
What I hope for
I hope your input is something you chose, not something others want to push on you, and not an ever-growing pile of junk. A lot of what creators write with AI these days is, frankly, filler. Good output starts with good input.
I hope you use this tool to make your input better and better.
What it does now, and what is next
Today it supports YouTube. What I want to build next is support for more video platforms, and, while recommending, to surface good videos from similar channels so you can discover channels you don't follow yet.
To see how it feels, open a real example: AI founder picks. To make your own, create one for free.