The only way to make that bearable for me is to either solve the meta problem which often isn't feasible, or turn on some mindless show that I've already seen and don't have to pay attention to enjoy. I find in particular The Simpsons seems to often hit that sweet spot between funny enough to be entertaining and mindless enough to not be distracting.
Whereas if I try and listen to a podcast or audiobook it requires too much focus and I miss the book or can't pay attention to that task I am working on, liking filling in a years worth of time sheets at the end of December.
10 lines of Python and some browser automation. You did mention "solving the meta problem", and this looks like it is a candidate for that.
I also think that whilst the algorithm is quite good at finding things I'd like, it's quite bad at contextualising them so music videos and tutorials get mixed up in playlists it autogenerates for me anyway.
I'll put documentaries or other informational content on as "noise in the background". If I'm trying to sleep or whatever, then each re-watch can start a few minutes later depending on how much I could recall/recognise.
That's a great point, and could very well explain showing me videos I've already watched. Fixing my other two complaints would require keeping track of how many times a video has been recommended, and consulting my watch later list when building the recommendations list. It seems safe to assume YT is not unaware of these possibilities, so they must have reasons for not adding these signals in.
If you look at the recommended content for most major services, they give a lot of space to "stuff you've consumed before". Anecdotally, this is true of Netflix (which tends to rank already-seen content below it's own-brand shows but above other third-party stuff); and it's especially true of Spotify, which is constantly pushing me (in a real variety of ways) mostly music that I've already listened to.
I think it's likely that, music video or not, most users on YouTube are watching the same videos over and over again - hence why the algorithm tries to hard to cater to those users.
From a cynical perspective, such an algorithm is also a "safe" bet - if you won't even binge videos YouTube knows you like (because you've seen them), it will probably be much harder to get you to watch lots of videos you might like. Users who like to see the same stuff over and over are probably giving the best return on YouTube algorithmic investment, and they might even be YouTube's most valuable users outside of content creators.
But I agree with larger point.
i thought this was a bug and I contacted Netflix about it. found out it isn't a bug, it's a feature. although i do not know anyone who would actually watch a 2hr movie the very next day after watching it in the first place. realistically their discovery algorithm is not really an algorithm, more like a static list of their own productions. i do wonder if the "like/dislike" buttons have any sort of effect, but it doesn't seem so.
> Spotify
i rarely use Spotify as a music discovery tool. i usually read reviews from various sites then I search Spotify for the exact album/song. but what i found is that Spotify can actually give me some good recommendation after i filled a playlist with songs that i like.
Kids.
Also, people who fall asleep watching TV. Some people share a profile.
1. I still don't want them showing up in recommendations. Recommendations are not a history view, if I'm listening to a song on repeat for 5 hours I don't need Youtube to recommend it to me. It has been recommended once very successfully, and if I need to find it again it should show up in my history[0], or at least it should be quarantined to its own recommendation section that I can toggle off. Even among people who obsessively revisit old videos, revisiting old videos and looking for new content are two separate categories of searches that should usually be handled differently imo.
2. The fact that I revisit certain series or listen to music on repeat for 5 hours does not mean that I want every video recommended to me that way -- ideally it should just be the variety clips and music videos. Youtube's algorithms are theoretically advanced enough to tell the difference between music and a programming tutorial or product review. So there shouldn't be repeat recommendations for me showing up for stuff like... one video in isolation out of an entire Lets Play playlist.
I feel like, Youtube knows that's not a music video: I know they know because they already put the video in the gaming section. They're just being obstinate and refusing to use that information in their recommendation engine, and refusing to ask the obvious question of "is a music video consumed differently than one video out of context from a Lets Play."
----
I don't think you're wrong, I think you're probably 100% correct, I think that Youtube kind of just optimizes to get people to click on things and repeat videos are a way of doing that, and the algorithm just does what it does. But I'm not sure I know anyone who likes that, I should be the target audience for that behavior, I obsessively rewatch stuff all the time, and I don't like that behavior, the algorithm doesn't feel like it's tuned to my preferences. It feels like just kind of mish-mash.
In theory, if it was optimized for me, then I shouldn't be losing track of music. But it's not like its reliable enough to use as a bookmarking system, if I have a favorite song on Youtube I like to listen to, then I can't trust recommendations to have it in front of me, I have to actually bookmark it or (nowadays) download it and save it to an offline playlist.
So if they're trying to please people like me, then... it's not working.
<ranton>Personally, I feel like every time I interact with Google products or try to dig into how they work or recommend things, I always feel like the designers/programmers are almost deliberately trying to make their own jobs harder. Like they think it's cheating or something to have a toggle that could let people turn off this behavior. They have to organically have an AI come to every decision or else it's not 'real' programming; it's very weird.
I don't understand the reasoning why this wouldn't be user-configurable behavior at the very least, why it wouldn't be something where we can apply human reasoning on top of the AI algorithm. If you work at Google, I promise that no customer is going to be mad at you or shame you or say that you're bad at programming if you give us really clear controls that let us just tell you what we want in a predictable way. No one is going to be offended if you just ask us if we want to see repeat videos, you don't have to intuit that, we can all tell you ourselves and we won't be mad that you asked.
I hate feeling like I have to do reinforcement training on a cat to get good Youtube recommendations just because some product designer is scared that I'll think less of them if they ever admit that they don't magically know my preferences without asking. Some people might want repeat videos, and some people might not want them, and this shouldn't be an existential problem for Youtube, I thought the whole point of this algorithm stuff was that I could have an individual experience that was customized to me. Instead we've got the worst of all worlds because Youtube thinks its a personal failure if they don't come up with one single, universal strategy for serving videos that makes everyone happy across the entire world.
----
[0]: Shoutout to NewPipe that allows sorting history by the number of views. History in general could be way better on pretty much platform, but even just that sort option is useful on its own.
Google's entire business model is based around customizing content to an individual level. Youtube's whole deal is that it theoretically understands what I want to watch.
But it doesn't. And it's really weird to me to even have a discussion about what the average user might or might not want out of a recommendation engine. I feel like that misses the point -- with all of the data Youtube collects on users down to what portions of every video they watch, why is it so hard for individuals to tell Youtube what kinds of videos we want recommended?
Why is it showing repeat recommendations to people who don't want them? Google's entire business from Youtube to advertising to its calendar features, it's all built around knowing what individuals want, and this is an instance where even when people are consciously trying to signal to Youtube that they don't want something, they can't, and the site just kind of ignores it. It's a failed outcome no matter how you look at it; the average person's recommendation tastes for repeat videos don't matter because Youtube isn't just the "trending videos" page, it has a section called "your recommendations".
Those recommendations should reflect the preferences of the individual using the site, that was the entire promise behind all of this invasive user tracking.
You just bought a new car? have some car ads.
You are a guy? Have some scammy dating site ads while you already are dating, as you get enganged, married and have more than three kids.
Doesn't matter that you have taken the time to click not interested / not relevant.
My guess the advertiser wasn't too careful and paid good money for pay-per-impression in those spots.
On YouTube I click this every time, marking "irrelevant" vs. "repetitive" ads. Since I've never seen a relevant ad.
Sometimes I even go "inappropriate" for casino ads (it feels like it's not just irrelevant but inappropriate to try to encourage me to gamble).
I've never once noticed Google take that into consideration. Marking an ad "repetitive" seems to encourage it to repeat.
Your answer to the above might be "never". Which is totally cool.
But I think lots of people like rewatching favourite content, and YouTube is no different in this regard.
They don't care about recommending you something new, they care about recommending you something that you would watch.
I honestly don't very often, and when I do it tends to be over a span of time long enough to forget many of the details. I do have some music videos on youtube that I have watched several times, and some instructional videos I have returned to occasionally.
So yeah probably its the weakest of my three whines.
Very rarely. Even more rarely do I re-watch movies less than a month later, but that doesn't stop Youtube from recommending things I watched the day prior.
If I got recommended something I watched a year ago, sure. But what youtube frequently does is just recommend all of the videos I watched the day prior.
I would be less annoyed if this wasn't such a recent development. Youtube recommendations used to be fantastic for me ~5 years ago. I would actually get recommended stuff I haven't watched, from creators I didn't know about, on topics that I have shown interest in. Now I mostly just get the same videos I have already watched or nearly the same video uploaded by another channel. Or I'll watch a single 15 second video about something I normally don't watch, and suddenly half of my recommendations are about that.
I've only seen it right at the start or end of the list.
There are videoed you want to watch and there a videos you want to want to watch.
There’s a quote about classics book being the books you want to have read.
I think YouTube is filled with interesting content but they always choose to recommend the same videos for weeks!
However, their "New to You" section is much better and I think it should be on the home page.
That being said, one would hope that Youtube would have better user segementation, but I can't imagine that is an easy problem.