Hacker News new | ask | show | jobs
by stickfigure 2150 days ago
I rely heavily on Youtube's recommendation engine and I think it's fantastic. Not sure what your feed looks like, but mine is full of trains and airplanes and machining videos and history pieces and technology breakdowns and... I wish I had a lot more time to watch.

If you're getting terrible content, I have to think you're the one to blame. You have the like and subscribe buttons, you can train the system. I did.

1 comments

You can train an obscure and undocumented ai. Sounds simple enough.
Poe's law seems to be in force here, I can't tell if you're serious.

Thumbs-up stuff you want to see more of, ignore or thumbs-down stuff you want to see less of. Subscribe to channels that seem to have a lot of stuff you like. It seems straightforward?

I'm not sure thumbs up/down matter in the algorithm. But that's the problem: we don't know.

I understand you can watch only one type of video, but, from what I,'ve seen, as soon as you watch/click on 1 political video, you get bombarded by it. Also sharing ip with other people seem to affect what you're being shown. I'm surprised that hn seems to think it has that much control over the recommendation algorithm.

It’s been shown again and again [1] that YouTube endlessly promoted conspiracy videos, fringe thinking, UFO “truth”, and other such garbage for wide swaths of its audience. They are a parasite on the way the algorithm works, feed by clicks.

It’s a bit like blaming the heroin addict who is offered it everywhere they go. Don’t blame the user when this is wide spread.

[1] https://mashable.com/article/youtube-conspiracy-theory-recom...

They promote what they asked for. Assigning it agency from its content would be fundamentally ignorant and/or disingenuous. It is the exact same thing as the Google testimonies where congress members were upset that negative content came up when they were searched for.
Who is they and what is the ask? YouTube serves this to you by default.
Unironically yes actually. The inputs are simplified and documentation wouldn't really help anyway given it is based on such a vast dataset it is less trying to drink from a firehose and more trying to drink the entire ocean at once.