Hacker News Reader

Comment by CrazyStat | Hacker News Reader

Youtube has more than that uploaded every single day.

YouTube has on average close to 4 million video uploads per day

And still can’t recommend me anything interesting outside of my chamber.

I wouldn't mind if it didn't keep recommending me videos I've already watched (they're showing the red bar! they know I've watched it!) or from channels I've already said no to or on topics I've frequently said no to or on topics that I watched one video on[1] and now YouTube thinks I want a full page of similar videos.

The algorithm is, for want of some better words, absolute cack.

[1] Even if I don't finish the video or say I don't like it or say "don't recommend channel", FFHS

Lol, I know the pain (who doesn’t). Disliking and removing from history works for me, at least in the short term. I put history right on the sidebar with unhook extension, I believe. Sometimes I clear whole pages from there to avoid spam.

What to do with videos I don’t want to dislike, idk. It keeps recommending watch-once-already-watched videos indefinitely.

I solve the problem by watching less and less on youtube. I think it's the solution they wanted.

I use the like buttons to show myself I've watched a video. It does not result in optimal video recommendations, but hey ho.

The (dis)like buttons are now completely pointless on Youtube, since they removed dislikes when people disliked the wrong things and made Susan (RIP) sad.

It feels like the Youtube algorithm is on drugs now. I get a lot of uninteresting recommendations of videos with no views from channels with no subscribers. A lot of recommendations of things I've already watched recently. A lot of recommendations of stuff that's extremely old (a lot of which I've already watched).

And these recommendations persist for ages. Somehow it must know after a while, after presenting me the same video 50 times that I'm not going to watch it. It's so tiresome. And god help me if I watch a few cat videos. My recs are going to be full of cats for weeks. I feel like I have to watch cat videos in incognito.

> when people disliked the wrong things and made Susan (RIP) sad.

That is wild.

The goal is maximum relevant ads shown per day, not maximum interesting videos discovered.

You aren’t the target market.

99% of my youtube viewing is via the subscriptions page

Just wait until your kid borrows your phone,one time, to go watch Minecraft videos. You will never escape.

> to go watch Minecraft videos

Yep, I do that occasionally and bloody hell, you're not wrong about the flood.

Or your in-laws watch your Netflix over their holiday visit. Now I can't tell if Netflix has generally worse quality content these days or if it's just recommending garbage to me based on what people have watched on my profile.

I think it's just pretty hard to recommend things outside of what they know you like.

You can log off and look at what's popular on your country's front page. I get a couple popular music videos, clash of clans, soccer, and a whole lot of clickbait/prank/you won't believe this/pikachu face thumbnail videos.

Won’t work without trying.

It wasn’t hard 10-15 years ago when it was at its peak. I suspect all it did was relaxing walk rules and simply presenting more diverse selection, which optimizes for surfing but deoptimizes for some modal group.

I also remember right-bar surfing techniques that don’t work anymore.

Front page is distilled garbage to me.

Part of it is also the creators optimizing for what the suggestion algorithm rewards. If you don't like the front page, you're looking for a needle in a haystack, because creators are rewarded very heavily for making the kind of content that lands on the front page and gets mass amounts of views.

Youtube doesn't have much of an incentive to show things that are not either popular in general, or that already worked on you in the past. There are some people that care strongly about that (me included), but it's a very small minority.

If the A/B test says distilled garbage is what hijacks the dopamine center the best, then you will be fed the A/B juice. And on average, people in your cohort will like it.

I remember the right-bar surfing, too, when you got new or related / relevant videos.

I seen multiple creators performing A/B tests on clickbait titles/thumbnails/pokatchu faces and everywhere conclusion was the same: it's so effective for their bottom line that even when they are not comfortable with it, they cannot afford to not do it.

Above note is the reason Google can't recommend you anything.

Consider momentarily the amount of data processing necessary to somehow recommend a relevant video from:

  ~14,000,000,000 videos on Youtube
  615 seconds (~10 minutes) mean length.

Which works out to:

  8,610,000,000,000 seconds
  143,500,000,000 minutes
  2,391,666,666 hours
  3,274,083 months
  272,840 years
  27,284 decades
  2,728 centuries
  273 millennia

Netflix simply attempting to provide somewhat relevant recommendations was a massive data crunching effort years ago, and even that was "only" the official movies and television of humanity. Data take from a previous post I made 9 months ago [1] and from this article [2] and this paper on Youtube data statistics [3].

[1] https://news.ycombinator.com/item?id=39421041

[2] "What We Discovered on ‘Deep YouTube’", https://www.theatlantic.com/technology/archive/2024/01/how-m...

[3] "Dialing for Videos: A Random Sample of YouTube", https://journalqd.org/article/view/4066/3766

It can't be that hard, once you develop a profile for a user, you just need to classify the incoming videos and cross reference their profile against the classifications.

I think that “just” might be doing some heavy lifting in that assertion.

Sure and classifying video is one of the classically hard things for computers to do. Really classifying in general.

Underrated reason for tiktok's success: shorter videos lend themselves way better to recommendation algorithms, because you have better data about what users want to see

Almost all my recommended videos have tens of thousands of views or more. I don't know for sure if the algorithm usually ignores 97% of videos, but it might as well be doing that. Doing that vastly reduces the number of options and means you have lots and lots of data for each video.

Turn off watch history in your google settings. Then remove all your favorites. (save them to a playlist first to keep them) This forces the algorithm to only consider what other people who have watched the video you have open are watching when building suggestions, and this makes the suggestions dramatically better.

At the top of the homepage there's a menu bar with categories, if you scroll all the way to try right of that there's a "new to you" option. Try it out :)

Youtube is one symptom that Google's business model is ethically and ecologically unsustainable. Process and store whatever nonsense to collect more user data.

Are there any recent stats on how many hours (maybe days now) that are uploaded per second?

I'm seeing a variety of places all saying 500 hours per minute, or 30,000x as much upload as clock time.

Reply on news.ycombinator.com