Totally agree. I've found the recommendation engines on all the major content platforms to be pretty annoying - it's the echo chamber, basically recommending stuff I already know I like. It feels like eating junk food: I enjoy it when I'm eating it, but ultimately I end up regretting it.
I agree, with one caveat: Spotify's algorithm seems less shitty than most. I've discovered a decent amount of fantastically good, fairly obscure music.
I can't shake the feeling the experience just isn't optimised for what we want as users. For quite a while Google Play had fantastic organic discovery based on previous likes, but then it just got taken... elsewhere.
I don't use Spotify, but on a similar note, my lone positive experience with any kind of recommendation algorithm has been with Apple Music/iTunes. Perhaps it's because I have used the platform for as long as it's been around, and have always been actively creating playlists, rating tracks/liking tracks within iTunes itself, thus ensuring their "genius" algorithm has sufficient information on me, but I have consistently managed to discover a lot of great bands/tracks/albums from it. Especially over the past year or two, it's pretty impressive.
Yeah their radio thing is my main way to find music these days. Find a song I like, hit radio on it, do some stuff till one stands out enough to like the song.
I do wish it'd stop trying to play Lostprophets on the odd occasion though. I don't think I've even listened to them even before the crimes but I don't want to enjoy a song and find out it financially supports a pedo.
I used to say that too, but it's been years. Spotify used to be absurdly impressive in finding obscure stuff I was interested in, but it just doesn't do that any longer.
It may be that they changed the algorithm to work better for people who are not me. It may be that the algorithm choked on its own feedback (once listening patterns influenced by Discover Weekly started influencing Discover Weekly). Or maybe it just plain ran out of stuff to recommend me, having picked all the low hanging fruit (from its understanding of my taste).
It's also a bit conspicuous that DW was at its best when Sander Dielemann and Aäron van den Oord were interns working on it. Both went on to write some pretty great machine learning papers, and both were snapped up by Google/Deepmind.
I'm not defending Algorithms but manual curation doesn't scale. There's hundreds of content-hours uploaded every minute on YouTube. YouTube would need tens of thousands of coolhunters constantly watching video all day constantly updating recommendation lists. If a user's feed was just a simple chronological list it would just constantly scroll all day. No one would have any time to watch videos as they'd just be scrolling through the list of videos.
That's not to say YouTube's recommendation algorithms are good or that everything needs to be an algorithmic feed.
>I'm not defending Algorithms but manual curation doesn't scale.
There are tons of employees at YouTube. A lot of moderators as well. Each one of them could easily discover tons of content within YouTube to publish on personal pick pages on the site. YouTube chooses to only feature a handful of creators at a time because that protects revenue streams, not because content moderation can't scale. People would literally jump to curate content for free on YT as well, as many playlists are made every second by users.
I'd love to see a real honest stat on how views are distributed across YT content, I'm pretty sure the relatively few items featured on their front page and trending/sponsored playlists get the lion share of all views annually, and everything else suffers in silence (relatively).
Sure, but like 99% of the stuff that's being uploaded every minute to YouTube as you mention isn't getting recommended to pretty much anyone, ever, anyway. The vast majority of videos uploaded to YouTube have abysmally small view counts/reach, and the bulk of actual views on the platform come from content regularly being uploaded by a small percentage of the overall user base. Same as pretty much any platform, just at an exponentially increased scale due to how enormous and popular YouTube is.
I also believe it's entirely possible for someone (it won't be YouTube, they are too invested/tied into their existing model) to successfully integrate manual curation at scale. All you have to do is monetize the curators, and then reverse the traditional ad model being used on the majority of the web. Just like how people are already driven to go and watch live streams from creators who oftentimes simply serve to be a more active, on-screen role of one of these kinds of curators or "coolhunters" or whatever...it's been shown that people are more than willing to donate via superchats or sign up for additional membership features. So what's stopping a platform from doing more targeted audience-specific advertisements?
People just don't listen. They have no clue that most of what is playing everywhere is sponsored ads, and content promoted by the people who upload it in pretty much every niche that can make money or make someone notable.
These sites have laid off a lot of their staff that once developed and maintained algorithms, and people were cheating through weak algos en masse back when sites were running wild, now the lazy solution is to sit back and collect ad money while just making bug fixes.
Agreed. Monetization and aid promotion for every account is what destroyed the social media world. Now nothing grows organically and feedback/comments/reviews are regularly reflective of popularity/favorability deceit.
> They have no clue that most of what is playing everywhere is sponsored ads, and content promoted by the people who upload it.
I'm always horrified at how bad people are at even recognizing an ad when they see one these days. I thought maybe it was just banner blindness, but I don't think many people even understand what an ad is.
Once I had someone argue that windows 10 didn't contain any ads at all and as proof they sent me a screenshot of their start menu which sure enough was filled with ads. It was struggle to get them to accept that all the icons for apps that weren't installed and services they didn't use were ads. I've seen people insist that Netflix didn't have ads just the full screen banners for whatever they were promoting you have to scroll past to get to your list/continue watching or the "suggestions" they throw up after a show ends. If it wasn't interrupting what they were watching every 10 minutes it didn't count to them.
For as much as advertising has found its way into everything, people seem to be getting worse at realizing when they're being manipulated.
for me algorithms are wrong far more often than human recommendations. it’s not even close.
i go to my local music store or friends and get almost consistently solid recs while youtube’s and spotify’s are just ridiculously bad.
same with movies, books, restaurants, etc… recs from people i know are just better. even better when the people are wonks, e.g., recs from actual film geeks rarely let me down.
and it’s a significant difference. it isn’t even close.
liek 6 months ago i completely switched to solely human recommendations and it’s drastically improved not only what i consume, but it’s been kinda rad to to find those kind of connections again.
Can't relate. I've spent a lot of time aggressively tuning my YouTube and TikTok recommendations using the "not interested" button and now 90% of the videos on my homepage are 10/10 bangers.
Try opening YouTube in incognito and watching the first 10 videos. You'll cringe so hard that your body will be twisted up like a pretzel.
Maybe frustration with YouTube somehow drives more engagement, so you'll go looking for new content and stay longer -- like how shops rearrange things to expose you to other goods and so increase sales.
Perhaps you need to change human behaviour so that most people don't rewatch things? (Speculation! Personally I prefer fresh content too.)
Maybe YouTube have found that rewatching items drives people to annoyance with the adverts and gets more revenue?
This is gonna sound pretty silly, but: You gotta scroll down!
The top recommendations on YouTube are often extremely "safe" in the sense that you'll feel bad about marking them as "not interested" because you already like the video and the uploader. This is likely a bug in their reward function, but if you skip the safe recommendations you'll start seeing more more intelligent recs such as videos that have the same presentation style as the ones you like except they were uploaded by unknown creators with 1k subs. Over time the algorithm will learn that you only like to rewatch certain kinds of videos (music videos in my case) and that you like to be surprised.
Note that refreshing the page doesn't help. It will keep giving you new equally-safe recs until you scroll at least 1 page down and force it to generate new recs.
>if you skip the safe recommendations you'll start seeing more more intelligent recs such as videos that have the same presentation style as the ones you like except they were uploaded by unknown creators with 1k subs. Over time the algorithm will learn that you only like to rewatch certain kinds of videos (music videos in my case) and that you like to be surprised.
Why oh why can't you just tell youtube that this is what you want, instead of some elaborate algorithm training dance?
Building an easily tunable recommendation algorithm has been on my bucket list for over a decade. I'm not aware of any serious implementations. I'm sure if you look at the issue trackers for any alternative client for YouTube, you'll find people who strive towards the same goal.
I have the same experience using Invidious, which just shows videos of all my subscriptions in the order they're posted, without using any of YouTube's recommendation algorithms. If I want to watch a video about a related topic, I'll just search for it.
And agree about the logged out view. It's not for me, but clearly has a huge audience.
Opening YouTube in incognito is a really depressing experience for me. Reminds me that I'm maybe not as in touch with the general population as I'd like to believe, which is a plus.
It also stokes my misanthropy, which is why I don't do it much.
Its depressing to say the least. Its bunch of click-bait outrage and sport trivia. And the god awful 'celebrities' fucking around and somehow have massive following.
Its a lens on gen pop, but i dont like looking through it.
Most likely because they don't know they're doing it. And most people couldn't explain what algorithm means, or what it means in the youtube/tiktok recommendation context.
My memory of the curated days were that it was a lot similar to what you scroll past now. Lots of niche stuff I didn't care for, but would pause because "someone" thought it was worthy.
They also had less to choose from, so it was easier to surface interesting things.
The years just prior to the explosion of these content algorithms were really magical.
Some time between P2P networks and piracy sites being shut down there was this amazing period of discovery. It seemed like it was mostly blogspot curated mega uploads of flac files. You could literally just curl huge, lossless,
obscure records that someone else found for you. It really felt like a beautiful flash in a pan. I’m sure a lot of people still have these files hanging around on hard drives but there’s no incentive to be the curator anymore.
There would be blogs of high quality vinyl and cd rips of
really thoughtfully curated clusters of types of music. I actually remember not being surprised when Kim Dotcom got arrested but being very bummed knowing the era was over.
Soulseek lives on as a weird vestige of this time, but the curation is not really there.
Those sites still existed last I checked, which was a few years ago. Once in a while I get nostalgic for something I grabbed on one of those sites around 2005, I'll run a search for it and find three new blogs posting pirate links to an eclectic mix of music they consider cool.
I remember reading that Google did something to derank them, so they may finally be dead, but they continued on for a long time past when I visited them daily for hipster cred.
2015-6 or so was the magical era for me. Intelligent enough to suggest you related content before you knew you needed but still early enough to not have to suffer through endless repetitive video spam for ad views.
A specific window sometime between 2000 and 2010. And a majority of the US wasn't.
I've had the same conversation with old hippies about record stores and how you just don't appreciate the music when you can just find it online.
The point is that old people have rose tinted glasses to when they were finding out new things for the first time regardless of how good or bad it was.
My own take for the time was that we were starved for good video content which is why we focused on the much smaller and more easily compressible audio files.
> Today, YouTube is known for its powerful recommendation algorithm: a system criticized for driving people to radical beliefs, conspiracies, and online echo chambers.
It's been trying to radicalize me into playing Minecraft and listening to K-Pop. I have no idea why.
Youtube puts certain content creators on the front page. Those creators get millions of views. Others pay for promotion on their uploaded content, they line the secondary pages, ranked based on how much they're willing to pay. Everyone else swims in the dead sea of ignore, unless they realize that they need to create and subtly promote their own website on TikTok & IG furiously, and embed their youtube videos into it.
Who knows if any social stats are real any more. THE END.
By taking YouTube's present zeitgeist of content creation as reference, this leaves out a big element of early YT: Broadcasting Yourself -- that social networking through informal short-form videos, spoofs, and replies that YouTube relinquished to other platforms over the years and is now hastily attempting to recapture.
You can still get a glance, starting at the Internet Archive...or by finding a subculture so niche and so dead that the algorithm can't even find anything recent and irrelevant to suggest (must be incognito to avoid personalized results): https://www.youtube.com/results?search_query=suche+emo+fruen...
From video responses to annotations to community captions, my, what a boon to user expression design-by-analytics has been!
I recommend pockettube browser addon, add your favorite creators channels into your own channels feeds, then you turn yt back into uncurated feeds. Love it.
I put my favorite car channels into a #Cars group and get unfiltered feed like an old fashioned dvr. And then if a video is interesting, click hashtags displayed for the videos for more of the same content. Find a new channel, add it to your group. In a hurry, show last 100-500 videos, sort by top, watch a few. Time to kill, sort by new, and scroll.
Works, prob the best solution until something better comes out.
Amazon experienced a similar transition in early 2000s:
> The editorial division, which dated back to Amazon’s earliest days, was composed of writers and editors who added a human touch to the Amazon home page and to the individual product pages...An algorithm called Amabot brought about the downfall of editorial. Amabot replaced the personable, handcrafted sections of the site with automatically generated recommendations in a standardized layout. The system handily won a series of tests and demonstrated it could sell as many products as the human editors.
Source: The Everything Store: Jeff Bezos and the Age of Amazon by Brad Stone
68 comments
[ 3.0 ms ] story [ 147 ms ] threadMaybe one day people will admit running your life through an algorithm isn't it.
I do wish it'd stop trying to play Lostprophets on the odd occasion though. I don't think I've even listened to them even before the crimes but I don't want to enjoy a song and find out it financially supports a pedo.
It may be that they changed the algorithm to work better for people who are not me. It may be that the algorithm choked on its own feedback (once listening patterns influenced by Discover Weekly started influencing Discover Weekly). Or maybe it just plain ran out of stuff to recommend me, having picked all the low hanging fruit (from its understanding of my taste).
It's also a bit conspicuous that DW was at its best when Sander Dielemann and Aäron van den Oord were interns working on it. Both went on to write some pretty great machine learning papers, and both were snapped up by Google/Deepmind.
That's not to say YouTube's recommendation algorithms are good or that everything needs to be an algorithmic feed.
The only reason algorithms are used is because of engagement.
If people don’t engage then they don’t watch as many ads or they eventually cancel their subscription due to lack of use.
So YouTube uses automation to shove their content down your throat.
There are tons of employees at YouTube. A lot of moderators as well. Each one of them could easily discover tons of content within YouTube to publish on personal pick pages on the site. YouTube chooses to only feature a handful of creators at a time because that protects revenue streams, not because content moderation can't scale. People would literally jump to curate content for free on YT as well, as many playlists are made every second by users.
I'd love to see a real honest stat on how views are distributed across YT content, I'm pretty sure the relatively few items featured on their front page and trending/sponsored playlists get the lion share of all views annually, and everything else suffers in silence (relatively).
I also believe it's entirely possible for someone (it won't be YouTube, they are too invested/tied into their existing model) to successfully integrate manual curation at scale. All you have to do is monetize the curators, and then reverse the traditional ad model being used on the majority of the web. Just like how people are already driven to go and watch live streams from creators who oftentimes simply serve to be a more active, on-screen role of one of these kinds of curators or "coolhunters" or whatever...it's been shown that people are more than willing to donate via superchats or sign up for additional membership features. So what's stopping a platform from doing more targeted audience-specific advertisements?
These sites have laid off a lot of their staff that once developed and maintained algorithms, and people were cheating through weak algos en masse back when sites were running wild, now the lazy solution is to sit back and collect ad money while just making bug fixes.
Agreed. Monetization and aid promotion for every account is what destroyed the social media world. Now nothing grows organically and feedback/comments/reviews are regularly reflective of popularity/favorability deceit.
I'm always horrified at how bad people are at even recognizing an ad when they see one these days. I thought maybe it was just banner blindness, but I don't think many people even understand what an ad is.
Once I had someone argue that windows 10 didn't contain any ads at all and as proof they sent me a screenshot of their start menu which sure enough was filled with ads. It was struggle to get them to accept that all the icons for apps that weren't installed and services they didn't use were ads. I've seen people insist that Netflix didn't have ads just the full screen banners for whatever they were promoting you have to scroll past to get to your list/continue watching or the "suggestions" they throw up after a show ends. If it wasn't interrupting what they were watching every 10 minutes it didn't count to them.
For as much as advertising has found its way into everything, people seem to be getting worse at realizing when they're being manipulated.
for me algorithms are wrong far more often than human recommendations. it’s not even close.
i go to my local music store or friends and get almost consistently solid recs while youtube’s and spotify’s are just ridiculously bad.
same with movies, books, restaurants, etc… recs from people i know are just better. even better when the people are wonks, e.g., recs from actual film geeks rarely let me down.
and it’s a significant difference. it isn’t even close.
liek 6 months ago i completely switched to solely human recommendations and it’s drastically improved not only what i consume, but it’s been kinda rad to to find those kind of connections again.
That's actually one of the reasons I've created https://digs.fm (think Goodreads but for music).
Try opening YouTube in incognito and watching the first 10 videos. You'll cringe so hard that your body will be twisted up like a pretzel.
Perhaps you need to change human behaviour so that most people don't rewatch things? (Speculation! Personally I prefer fresh content too.)
Maybe YouTube have found that rewatching items drives people to annoyance with the adverts and gets more revenue?
The top recommendations on YouTube are often extremely "safe" in the sense that you'll feel bad about marking them as "not interested" because you already like the video and the uploader. This is likely a bug in their reward function, but if you skip the safe recommendations you'll start seeing more more intelligent recs such as videos that have the same presentation style as the ones you like except they were uploaded by unknown creators with 1k subs. Over time the algorithm will learn that you only like to rewatch certain kinds of videos (music videos in my case) and that you like to be surprised.
Note that refreshing the page doesn't help. It will keep giving you new equally-safe recs until you scroll at least 1 page down and force it to generate new recs.
Why oh why can't you just tell youtube that this is what you want, instead of some elaborate algorithm training dance?
And agree about the logged out view. It's not for me, but clearly has a huge audience.
It also stokes my misanthropy, which is why I don't do it much.
Its a lens on gen pop, but i dont like looking through it.
Apart from a few geeks, almost everyone I know of says exactly that.
They also had less to choose from, so it was easier to surface interesting things.
Some time between P2P networks and piracy sites being shut down there was this amazing period of discovery. It seemed like it was mostly blogspot curated mega uploads of flac files. You could literally just curl huge, lossless, obscure records that someone else found for you. It really felt like a beautiful flash in a pan. I’m sure a lot of people still have these files hanging around on hard drives but there’s no incentive to be the curator anymore.
There would be blogs of high quality vinyl and cd rips of really thoughtfully curated clusters of types of music. I actually remember not being surprised when Kim Dotcom got arrested but being very bummed knowing the era was over.
Soulseek lives on as a weird vestige of this time, but the curation is not really there.
I remember reading that Google did something to derank them, so they may finally be dead, but they continued on for a long time past when I visited them daily for hipster cred.
Perhaps you could also infer that I used linked in during that time, or maybe even instagram! Google! Software engineer!
Just for the record all I’m saying is the blogs of this particular era were very interesting to me and I miss them.
I've had the same conversation with old hippies about record stores and how you just don't appreciate the music when you can just find it online.
The point is that old people have rose tinted glasses to when they were finding out new things for the first time regardless of how good or bad it was.
My own take for the time was that we were starved for good video content which is why we focused on the much smaller and more easily compressible audio files.
It's been trying to radicalize me into playing Minecraft and listening to K-Pop. I have no idea why.
Who knows if any social stats are real any more. THE END.
You can still get a glance, starting at the Internet Archive...or by finding a subculture so niche and so dead that the algorithm can't even find anything recent and irrelevant to suggest (must be incognito to avoid personalized results): https://www.youtube.com/results?search_query=suche+emo+fruen...
From video responses to annotations to community captions, my, what a boon to user expression design-by-analytics has been!
I put my favorite car channels into a #Cars group and get unfiltered feed like an old fashioned dvr. And then if a video is interesting, click hashtags displayed for the videos for more of the same content. Find a new channel, add it to your group. In a hurry, show last 100-500 videos, sort by top, watch a few. Time to kill, sort by new, and scroll.
Works, prob the best solution until something better comes out.
https://yousub.info/
> The editorial division, which dated back to Amazon’s earliest days, was composed of writers and editors who added a human touch to the Amazon home page and to the individual product pages...An algorithm called Amabot brought about the downfall of editorial. Amabot replaced the personable, handcrafted sections of the site with automatically generated recommendations in a standardized layout. The system handily won a series of tests and demonstrated it could sell as many products as the human editors.
Source: The Everything Store: Jeff Bezos and the Age of Amazon by Brad Stone