this is an awesome idea, for a handful of different reasons. "A feed of the present". The post got my attention because of the domain name. Perhaps including your description "A feed of a present" would get more deserved attention
PSA: Watch in an private/incognito tab/window. If you are currently logged into your google account, this WILL pollute your watched history: https://www.youtube.com/feed/history
they probably log a 'shadow view' but recommendations seem to be directly based on your watch history since your recommendations are lost when you completely clear it.
The recommendations will mostly be derived from the videos you have saved in playlists. And the front page will always shows Youtube Mixes about the last song you were listening to, at the time of turning off the watch history.
Do a significant amt of YouTube users use playlists? I’ve always assumed no. Unless perhaps they are doing it for music since you’re talking about YouTube Mix. And songs.
Because a lot of people care of recommendations? If you watch news on yt then maybe recommendations are garbage but if you watch videos related to some niche activity or music then recommendations are a godsend.
Some like me actually appreciate the youtube recomendations based on the previously watched content, but need to be extra careful when watching some kind of content that is likely to be weighted a lot by "power users". Example, I do not follow videogames, but I do enjoy watching speedruns of old games ocassionally, so I need to watch it on incognito mode so I won't have my recommendations flooded with videogame videos.
Or otherwise paid for (which I think still counts as 'ads'), or tuned towards engagement.
I think quite a lot has to do with machine learning working out that if you panic the human animal they pay attention to threats, and therefore to maximize engagement 'if it bleeds, it leads' (old newspaper maxim). Newspaper editors can (and might not) automatically apply a social-benefit heuristic or sense of social shame (not wanting to be a 'muckraker' or troublemaker), and machine learning may not even start with such a concept.
If engagement was maximized by turning viewers into cannibalistic humanoid underground dwellers (CHUDs), machine learning would simply make note of that and run as hard as it could in that direction, since it doesn't have a larger context in mind unless programmed to do so.
(Such a larger context is actually sort of controversial: lot of people demonize the very concept of social justice, and without it you get these hacks to maximize engagement by tapping into really unmanageable human/animal behaviors)
I'm currently unaware of any recommendation engine that's worth acknowledging.
Netflix, Amazon, Youtube, PornHub, etc... they're all accomplishing little more than "similar to the one, and only one item you last saw", with dramatic shifts in "profiling" from one or two videos.
Actually, Netflix acknowledges this and splits the recommendation into "because you watched X..", so at least it covers a greater range (eg last 5 things seen)
I'm damned sure they could be much more useful if they would let me tell them what I like, by implementing rating systems that are worth using (e.g. the ability to browse and edit previous ratings in a sane fashion)
but user-useful recommendation is not the actual goal, so really its just that our metrics are wrong. It's probably great according to view counts.
The only one I think is good is when I build radio stations off custom playlists on Google play music. Generally speaking, most of the songs are good, and like 40% will get added to the playlist as well.
I don’t use this, or Spotify, but those are the only two systems I’ve heard people give praise for the recommendation engine — I suspect that its because music playlists are almost equivalent to a rating system.
That is, the user is capable of efficiently informing the engine of their taste, and there’s significant incentive for the user to consistently re-evaluate their ratings (playlists), so it can be trusted as up to date.
Another very important aspect is that playlists are useful enough to the user that they actually want to maintain it.
For example, amazon, netflix and pornhub all have rating systems, but they’re not at all useful. The interface isn’t useful enough for reviewing and reflecting on, its not comprehensive enough to keep as a primary list (because it only covers what they offer, which is very limited) and there’s of course no impact on the recommendation engine (because the rating systems are not worth using; chicken and egg). No sane person would touch the things (beyond “upvoting”, which isn’t significantly related to taste)
Imo ratings are absolutely vital to useful reccomendation, but they’ve been totally neglected
The radio stations off custom playlists are indeed pretty good, but the recommended ones on the home page are often horrendous and very short (songs will basically get repeated after less than 1 hour).
last.fm is great for recommendations after you've been scrobbling for awhile. I've picked up so many artists from checking out recommended every once in awhile.
That's the bottom line right there: thinking that recommendations are useful to YOU is naive and missing fundamental things about the experience.
That said, you can indeed tell them what you like but by use of negative space. When you get bombarded with obviously horrible recommendations, do the two-step process of clicking 'Not Interested' (if possible without even watching the video, or you can check it in incognito mode, assuming they're not watching that even more closely) and then 'tell us why', and respond 'I'm not interested in this channel: "undesirable video maker".
That assumes you can be sure you want to nuke the channels and subjects in question, but when it's clickbait channels and/or alt-right propaganda it's generally easy to identify and not get wrong. I'm sure the same would be true for leftwing propaganda, but the stuff I don't want pushed on me has a whole language and lexicon that's easily recognizable by video title, channel title and attempted clickbait image. If stuff trips my sensors on those grounds, I'm generally comfortable nuking it unseen.
On this subject, clicking through to some of these videos in incognito gave me what I guess must be the "default" suggestions - which included not just beauty tips spam but far-right speakers and loyalist marches in Belfast. I'm guessing the latter is related to the date but it's not something I'd expect to be the default! Watching even one video then made it more normal, even when trying to generate recommendations based on the no view videos.
Others are suggesting practical reasons for having watch history enabled, but I would think that the vast majority have it enabled simply because that is the default.
Oh, you think you can actually disable it? its just an UI setting, Google keeps tracking you anyway. Go ahead and try it, disabling does nothing to the recommendations.
history augments memory when searching for specific things one remembers. having it polluted by random videos embedded on third party might not be great. history also drives recommendations, so again having random video views will ruin coherence of the home page suggestions
I wish you hadn't told me that. I thought I saw something suspicious when I resumed switching and yep, that's a bunch of topless elementary schoolers at a pool party. I suppose it doesn't technically violate YouTube's policies but jesus I did not want to see that.
You can click X on them and quickly remove them. If there's truly way too many (you left it open for hours), you can go to your Google's Activity page, filter youtube and delete everything from today with a button.
When I go to YouTube on a fresh device without being logged in, it's a pile of steaming clickbait and pop-internet-culture garbage. On my account, by contrast, YouTube is full of mostly great recommendations of high-quality content and I can pretty much always find several new and interesting things to watch should I feel like it.
(I watch stuff like Kurzgesagt, Smarter every day, AvE, Rick Beato, Today I found out, Wisecrack, Wendover Productions, Practical Engineering, Vox, Crash Course, SciShow... that's just from browsing my current recommendations. I would guess none of that shows up for fresh accounts)
Clean it of trash and you will receive less trash recommendations. Fully cleaning isn't effective as the average video quality is lower than most HN public would like to watch. Having mostly good videos you like makes it recommend more similar content to you.
It's not just average video quality, it's that (as far as anyone can tell), the algorithm optimizes for expected value of total watch time, as opposed to optimizing for just the likelihood of watching the next video. Basically, the recommendations behave as though they "hope" that you will go down a conspiracy or outrage rabbit hole and binge 12 hours of garbage.
For me it correctly "hopes" I'll go down science rabbit holes, which I do. I'm similar to the person above, I watch a lot of educational content, and I also get suggested great content.
Right now I see the follow up to the guy who build his own VGA card, video of someone building a camera that can see wifi, and a PBS video on the "quantum internet". All great suggestions.
The problem I have is that my recommendations are usually full of stuff I'm already subscribed to. Discovering new things is really hard and usually just happens by reading about it elsewhere.
Also consider using Firefox and enabling `privacy.firstparty.isolate`, which will separate the cookies for third-party embeds from their own domain, thereby preventing this (as the embed doesn’t see you as logged-in).
And it makes me wonder about using similar approaches to break down the echo chambers we find ourselves in. We have a perception of what's normal based on what we see, but what we see is based on what we're already exposed to and what we ourselves do. Randomly seeing what a bunch of other people did this week? Great for that.
I also saw someone rave about "Donut" this week - schedules random 1-on-1's with people in your company to help with cross-pollination and bigger picture context. Chat Roulette and what a dumpster fire that is comes to mind, but I wonder if a LinkedIn-based service of a similar nature would be good just to learn about other companies, other corporate cultures, etc..
Your comment reminded me of interviewing.io and their strapline "chat roulette without the dicks"
Similar idea but for engineering interviews and practice interviews
I work at a dumpster fire of a company, but Donut Buddies still shines through with benefits to me. I don't even think others are aware of its true value. I often will make cross-functional inferences about company level things, or will bring up strategy from a different part of the company in discussions, and I have other engineers kind of just stare at me and say ".... how do you know that?"
"from, get this, talking to my colleagues on a regular basis"
Does this respect the 'unlisted' setting for a video? I recently uploaded some videos and set them all to unlisted, and yet some of them received views despite me not viewing them or handing out a link. I meant to dig into that more but forgot after getting distracted. Can unlisted videos be found by a program like this, which I assume is using the API?
We [0] index YouTube actively and see way over 5.5B videos [1] at this point. We catch a lot of unlisted videos and we did try to figure out how is that possible in the past.
It seems that a lot of users will upload video which is by default published with the default settings and thus is visible from the outside. Even if they change the settings fairly quickly, automated systems like ours will already know about the existence of that video.
There could be other reasons but this seems the most likely, especially as a video that is being uploaded can be published fairly swiftly.
It sounds like you are aware that you are scraping videos that are later re-labeled as "unlisted", but you don't mention what you do to mitigate this problem.
Even if it may not be illegal, at the very least it would seem un-ethical to link to private videos like this, and it would seem trivial for you to "re-scrape" your database every now and then to check whether any existing videos have changed from listed -> unlisted, and if they have, remove them.
This logic would require them to re-scrape every video forever, which is unreasonable.
I think a better approach for everyone involved would be to only store references to videos which were posted more than x minutes ago. I'm not sure if they have that information when scraping though.
>It seems that a lot of users will upload video which is by default published [and then they change it to private] //
So to avoid that sort of unexpected public-ing (ie publishing) only one extra scrape would be needed. Or, if they knew the period over which the setting was normally changed then they could just delay the scrape until most would have already been changed.
I imagine though, in part, the 'fun' is catching inadvertent publication and morality is no t considered.
It actually has nothing to do with "fun". As I mentioned in my other comment, we don't expose our database publicly and nobody but us can see that a video is unlisted.
It would beat the purpose of our service would we delay our identification, and it would actually require some significant engineering efforts in order to introduce such capabilities into our system with significant economical impact on our business.
Has "unlisted" ever been known to mean "private"? I never assumed it was - rather it was just a video that would not appear in searches or recommendations on YouTube.
This is... fascinating on a very personal level. I've never been a "YouTube" guy; I'd rather skim/read an article than watch a video. I've never binged, never clicked-clicked-clicked my night away on Youtube, and generally when sent a 17 minute video tutorial, ask/search if there's a 30 seconds writeup.
But this... this is mesmerizing. As cheesy as premise may be, you do feel a little like an outsider voyeur - not in a perverse sense, but in the having-no-expectations-or-context sense. Each video proves a gem, and timing is right. And knowing that you may be the only person who has ever seen it just adds to mystique... absolutely brilliant! :O
It reminds me of the Adult Swim show "Robot Chicken". Not exactly in content but in format and style. Just random things, slices of life, from people across the world.
In the same vein of avoiding bubbles, I browse reddit by 'Top Of The Hour'. Filtrated enough to be decent quality, very fresh content, and not yet subverted by bubble affiliation or mind hiveing.
I'd like a back button, so the stream pause/resume isn't such a high cognitive load high-stakes high-regret "oh, that looks ... drat, too late" decision. Or perhaps a fade transition?
A somewhat similar "sampling" experience is searching for a high-frequency term like "a", and filtering for an upload date of last hour or day[1]. Non-English terms and time of day bias on geography.
This is the same search method I use to continuously play back YT videos on a dedicated small screen connected to a raspberry pi in my apartment. Definitely shows an amazing cross section through humanity. The great thing about the camera file name is it’s language and region agnostic. So you really get everything.
That seriously defeats the purpose of the site: unflitered discovery, nonjudgement, kismet, etc. Our global culture is deviating towards the norm because our tech encourages that behavior. I love it for what it is.
The entire value here, for me at least, is a) randomness an b) fact they haven't been watched before.
There's so much curatrd stuff (which is great), and all tech companies are trying so hard to send me what they think I'll like / agree with, it's refreshing to step out of that box.
Edit: apologies if, perchance, I missed some subtle sarcasm btw... You never know on them interwebs
It's a video of a woman reciting a poem that she wrote for her eldest son that speaks of her love for her son and her wish that he would get "off the streets". Emotional, honest, real. YouTube like I've never experienced. Brilliant.
This project is fantastic and has potential in bringing people closer together. I just watched a man propose to his girlfriend in another language. (I'm pretty sure she said yes!)
I found some obscure music video that had like ~200 likes with no downvotes but it didn't lasted long... not idea if it's bots that downvote stuff or just weird people (hate the word "haters" but that's what I mean)
It's when you change upvotes and downvotes so that the ration stays similar, but now you can't consistently find if your downvoting bot had any effect.
This is a great find and a very somber video. It brings me a smile to know that this lady will wake up to her video having thousands of views and hundreds of likes, as well as many lovely comments.
253 comments
[ 3.0 ms ] story [ 270 ms ] threadPSA: Watch in an private/incognito tab/window. If you are currently logged into your google account, this WILL pollute your watched history: https://www.youtube.com/feed/history
I think quite a lot has to do with machine learning working out that if you panic the human animal they pay attention to threats, and therefore to maximize engagement 'if it bleeds, it leads' (old newspaper maxim). Newspaper editors can (and might not) automatically apply a social-benefit heuristic or sense of social shame (not wanting to be a 'muckraker' or troublemaker), and machine learning may not even start with such a concept.
If engagement was maximized by turning viewers into cannibalistic humanoid underground dwellers (CHUDs), machine learning would simply make note of that and run as hard as it could in that direction, since it doesn't have a larger context in mind unless programmed to do so.
(Such a larger context is actually sort of controversial: lot of people demonize the very concept of social justice, and without it you get these hacks to maximize engagement by tapping into really unmanageable human/animal behaviors)
Netflix, Amazon, Youtube, PornHub, etc... they're all accomplishing little more than "similar to the one, and only one item you last saw", with dramatic shifts in "profiling" from one or two videos.
Actually, Netflix acknowledges this and splits the recommendation into "because you watched X..", so at least it covers a greater range (eg last 5 things seen)
I'm damned sure they could be much more useful if they would let me tell them what I like, by implementing rating systems that are worth using (e.g. the ability to browse and edit previous ratings in a sane fashion)
but user-useful recommendation is not the actual goal, so really its just that our metrics are wrong. It's probably great according to view counts.
That is, the user is capable of efficiently informing the engine of their taste, and there’s significant incentive for the user to consistently re-evaluate their ratings (playlists), so it can be trusted as up to date.
Another very important aspect is that playlists are useful enough to the user that they actually want to maintain it.
For example, amazon, netflix and pornhub all have rating systems, but they’re not at all useful. The interface isn’t useful enough for reviewing and reflecting on, its not comprehensive enough to keep as a primary list (because it only covers what they offer, which is very limited) and there’s of course no impact on the recommendation engine (because the rating systems are not worth using; chicken and egg). No sane person would touch the things (beyond “upvoting”, which isn’t significantly related to taste)
Imo ratings are absolutely vital to useful reccomendation, but they’ve been totally neglected
That said, you can indeed tell them what you like but by use of negative space. When you get bombarded with obviously horrible recommendations, do the two-step process of clicking 'Not Interested' (if possible without even watching the video, or you can check it in incognito mode, assuming they're not watching that even more closely) and then 'tell us why', and respond 'I'm not interested in this channel: "undesirable video maker".
That assumes you can be sure you want to nuke the channels and subjects in question, but when it's clickbait channels and/or alt-right propaganda it's generally easy to identify and not get wrong. I'm sure the same would be true for leftwing propaganda, but the stuff I don't want pushed on me has a whole language and lexicon that's easily recognizable by video title, channel title and attempted clickbait image. If stuff trips my sensors on those grounds, I'm generally comfortable nuking it unseen.
Good idea to do a full reset periodically anyways to reset your recommendations.
When I go to YouTube on a fresh device without being logged in, it's a pile of steaming clickbait and pop-internet-culture garbage. On my account, by contrast, YouTube is full of mostly great recommendations of high-quality content and I can pretty much always find several new and interesting things to watch should I feel like it.
(I watch stuff like Kurzgesagt, Smarter every day, AvE, Rick Beato, Today I found out, Wisecrack, Wendover Productions, Practical Engineering, Vox, Crash Course, SciShow... that's just from browsing my current recommendations. I would guess none of that shows up for fresh accounts)
Right now I see the follow up to the guy who build his own VGA card, video of someone building a camera that can see wifi, and a PBS video on the "quantum internet". All great suggestions.
Wow. It's a fascinating look at what likely makes up a large majority of YouTube content that I would otherwise never come into contact with.
I also love how the creator packaged it up as a though an alien visitor was using YouTube to sample our civilization. This is the 99%.
I also saw someone rave about "Donut" this week - schedules random 1-on-1's with people in your company to help with cross-pollination and bigger picture context. Chat Roulette and what a dumpster fire that is comes to mind, but I wonder if a LinkedIn-based service of a similar nature would be good just to learn about other companies, other corporate cultures, etc..
"from, get this, talking to my colleagues on a regular basis"
It seems that a lot of users will upload video which is by default published with the default settings and thus is visible from the outside. Even if they change the settings fairly quickly, automated systems like ours will already know about the existence of that video.
There could be other reasons but this seems the most likely, especially as a video that is being uploaded can be published fairly swiftly.
[0] https://pex.com
[1] https://blog.pex.com/what-content-dominates-on-youtube-39081...
Even if it may not be illegal, at the very least it would seem un-ethical to link to private videos like this, and it would seem trivial for you to "re-scrape" your database every now and then to check whether any existing videos have changed from listed -> unlisted, and if they have, remove them.
I think a better approach for everyone involved would be to only store references to videos which were posted more than x minutes ago. I'm not sure if they have that information when scraping though.
>It seems that a lot of users will upload video which is by default published [and then they change it to private] //
So to avoid that sort of unexpected public-ing (ie publishing) only one extra scrape would be needed. Or, if they knew the period over which the setting was normally changed then they could just delay the scrape until most would have already been changed.
I imagine though, in part, the 'fun' is catching inadvertent publication and morality is no t considered.
It would beat the purpose of our service would we delay our identification, and it would actually require some significant engineering efforts in order to introduce such capabilities into our system with significant economical impact on our business.
Also I don't believe unlisted videos are considered to be private. There is a private setting which disallows for public to see such a video.
And finally, it's not very trivial to touch 5.5 billion videos often enough to see if any of those became unlisted.
But this... this is mesmerizing. As cheesy as premise may be, you do feel a little like an outsider voyeur - not in a perverse sense, but in the having-no-expectations-or-context sense. Each video proves a gem, and timing is right. And knowing that you may be the only person who has ever seen it just adds to mystique... absolutely brilliant! :O
https://i.imgur.com/QE9U5tJ.jpg
In the same vein of avoiding bubbles, I browse reddit by 'Top Of The Hour'. Filtrated enough to be decent quality, very fresh content, and not yet subverted by bubble affiliation or mind hiveing.
[1] "a"/"last hour": https://www.youtube.com/results?search_query=a&sp=EgIIAQ%253...
https://www.reddit.com/r/IMGXXXX
0: That was the polite version.
1: http://youtube-dl.org/
The entire value here, for me at least, is a) randomness an b) fact they haven't been watched before.
There's so much curatrd stuff (which is great), and all tech companies are trying so hard to send me what they think I'll like / agree with, it's refreshing to step out of that box.
Edit: apologies if, perchance, I missed some subtle sarcasm btw... You never know on them interwebs
https://www.youtube.com/watch?v=LblUWrJXfSA
https://youtu.be/1rvPbeHjzlk
It's a video of a woman reciting a poem that she wrote for her eldest son that speaks of her love for her son and her wish that he would get "off the streets". Emotional, honest, real. YouTube like I've never experienced. Brilliant.
https://bruceworkman.com/about
Nice to see you completely ignored my advice though :P
I'd say that's why. To deter dislike bots