proxying youtube

today I thought writing a quick project, a youtube proxy server, as in, you browse localhost:<PORT> and youtube comes in the response.

this is not rocket science as proxy servers have around for a long time.

I thought it'd be interesting to code it in userland, as opposed to "systemland".

And 50 lines of code and some minor hurdles later I see youtube "running" in localhost.

Although youtube didn't just work as usual since the videos don't actually come from youtube.com, but from googlevideo.com instead. And my browser, expectably, enforces CORS and forbids any requests to it.

At that point I started to think of ways to somehow proxy googlevideo.com too. But the solutions are not at all trivial.

Then I thought what was the payoff of all this. I tried to proxy serve youtube out of curiosity, and sure thing, you can do it.

But what problem would proxying youtube solve? Maybe I should think in a fuller way what are the problems I have with youtube.

One issue I have is the exposure, discoverability. To explain it, let's say I have been watching a very, very big amount of videos as of today.

Personally I would expect youtube to understand very well by now what my tastes are, what do I want to watch and what I do NOT want to watch.

Notice that I am very black and white, and I do not have much interest in watching certain types of videos.

It could be true that if my expectations of how youtube should work became reality then youtube recommendations would become polarizing or echo chambering.

But that is my decision though, and the problem with youtube is that it's seemingly forcing a single recommendations algorithm onto everyone.

Some people are more open minded and want to watch EVERYTHING, and a lot of people don't.

But users aren't deciding what they should get recommended. Youtube is making that decision for them. And it sure feels like it's trying to maximize ad revenue.

I for one don't give two flying fucks about pranks or diva youtubers. Yet youtube is adamant in presenting some of these to me.

Now, trying to come up with a solution for this is really non trivial. It would definitely require some youtube mining, or some kind of network so as to not get rate limited when mining, and even then you still have to think of how a good recommendation system would work.

I think the implementation of all that would be too much for me (time and skill wise). But I think it's fun to at least try to outline how recommendations could work.

I would very much prefer that when youtube recommended something, at least it has some number of confidence meaning how much would I like that video, so at least I know what to expect.

It should also have some indicators like what is the mood of the video. As in, sometimes I watch youtube in the mood of learning, like programming videos, but most of the time I watch to get entertained.

These ideas are just brainstorms and could be terrible on reproduction, but I'd like to hear what ideas can some of the people here can come up with.