Hacker News new | past | comments | ask | show | jobs | submit login

I agree with this. Their ranker is really good and has improved dramatically the more I watch and like. And it's not even fully garbage content, a good amount of it is interesting, useful, inspirational and educational. A rabbit hole of really excellent videos from a professional dog trainer caused me a detour of 20 minutes and pushed me over the edge to create the original tweet. So one side of me is not even mad, but the other side of me is like wait where did 2 hours go? And do I really want this?



I always wonder what Bytedance is doing differently compared with other social recommendation tools (aka why does no one rave about Twitter's suggested follows).

My sneaking suspicion is product management designed a product that is literally perfect for a recommendation engine.


It really does seem like a lot of the magic here is simply the format. The video lengths are a perfect balance between short enough to be addictive, and long enough to enable content more varied and interesting than just short gags.

Instagram and Facebook get bland real quick. Youtube/Netflix videos are often too long to be really addictive - "just one more" isn't as persuasive if "one more" takes 10 min+. With shorter skits like Key & Peele on Youtube you do get a stronger sense of that addictiveness. Twitter is mostly text and much less expressive, and even influencers treat it mostly as a communication/announcement platform, not a content platform.

Vine was hugely popular when it was shut down by Twitter, and I think it'll go down in history as one of the largest corporate mistakes every made. They were so close on the format - they just needed to allow slightly longer videos (which they did with Twitter message lengths around they same time they shut down Vine!!!).

I'm sure Bytedance's secret recommendation sauce is good, but I think much of their success is simply thanks to them being the first competent executor of the next major social media content format.


> Vine was hugely popular when it was shut down by Twitter, and I think it'll go down in history as one of the largest corporate mistakes every made. They were so close on the format - they just needed to allow slightly longer videos (which they did with Twitter message lengths around they same time they shut down Vine!!!).

Vine's top creators "unionized" and started demanding to be paid a cut. Vine had to be destroyed to defend the principle of not paying social media contributors.


That's somewhat ironic since TikTok pays certain content creators, although apparently not well.


https://youtu.be/jAZapFzpP64

Worth the full watch on video media money. But tldw; TikTok splits a pot of money over the views each video gets. The pot doesn't grow with the revenue they make or total views, so it's pretty exploitative.


I found the opposite. I took a look at TikTok, but the videos are too short to get into anything interesting. There's lot of interesting stuff on YouTube, however. Some of them could be shortened somewhat, but not down to a few minutes.


TikTok has recently introduced 10-minute long videos. Although, I have yet to come across a video that is >3 minutes, or a creator that uses that length of video.


What about Instagram Reels? Is that bland too?


The content creators aren’t there. Most of the good stuff is ripped directly from Tiktok.

Also their recommendations aren’t as good but that could just be a side effect of the above.


Other platforms only look for similar content to what you've already seen. They don't take novelty into account.

If you like one picture of a dog on Instagram it'll just show you more dogs and you get bored. Watch a few YouTube videos about one topic and the algorithm gets fixated on it.

YouTube is so bad now that if you use it to listen to music the autoplay gets into an infinite loop of songs that sound near identical, it's torture if you don't already know what you want to see/hear.

In short the usual algorithms are tailored to find similar content only, rather than finding novel content several degrees of separation away from what the user has expressed interest in.


It also seems very easy to take recommendation into your own hands on TikTok with long press -> “not interested”, which is very accessible. Once I started to get quite a bit of political / cultural war crap in my feed; I “not interested” a few of them, the feed improved markedly almost immediately. With YouTube I guess you can sort of influence your recommendations by digging into your watch history and deleting a bunch of stuff, but it’s tedious and ineffective. “Not interested” on YouTube front page seems useless, I still get pushed the same repetitive stuff, maybe from another set of channels.


TikTok and HN (to a lesser degree) are the only places that don’t force feed me racial culture war stuff.


> with long press -> “not interested”, which is very accessible

First time I heard about it. If it's hidden behind an undiscoverable gesture/tap, it's not accessible.


IIRC it’s among the first things they onboard you about. It’s also in the share menu, you should have seen it if you ever tried to download or share a video, or create a duet, or something. Long press is just an easy way to bring it up. Given the percentage of people commenting on TikTok threads who’ve never used it, “first time I heard about it” doesn’t mean much.


"Don't recommend channel" is very effective for me to ignore creators I just don't care about or content I'm at moment not interested in.


YouTube recommendations used to be quite good for a brief amount of time when google brain originally took over it I think? Now it’s a mess but it’s clearly due to exec meddling. like their absolutely product destroying migration towards videos 8 minutes or longer; most original content is now fully shit because they are filling it up with garbage to pad the time and release in a cadence. All stupid rules imposed by YouTube, which just takes creativity out of people.

It’s also possible any recommendation system can only stay pure for a few years until everyone from both sides is gaming them so badly for money it just cannot work anymore.


I was trying to show a friend some features of a keyboard I just bought that I’m pretty excited about. It’s still in the mail, so we pull up YouTube on the nearest screen and type in the name of the board. The first and most popular video that came up was 14 minutes long, but looked high quality. As we skim through it, we realize that it’s mostly this guy speaking ad nauseam about himself with very few significant shots of the keyboard. We had to go back to search and go all the way to the long tail keyword-wise to find a video that gave us what we wanted. Yeah, YouTube search isn’t what it used to be, and my experience tracks with your suggestion that it’s executive meddling.

If that’s what YouTube wants to be now, that’s fine. My question is now where do I go to find what YouTube used to be (TikTok, maybe)? It’s ironic too, because it’s like they don’t realize that as a millennial I was drawn to YouTube because it _wasn’t_ TV.


On the other hand, if I ever fall alseep in front of YouTube with the autoplay on, I will more often than not wake up in front of either Tom Scott's unedited video about sending garlic bread to space[1] or Micheal from Vsauce reciting primes for 3 hours[2] (which tends to result in some pretty interesting dreams tho)

[1]: https://www.youtube.com/watch?v=YKAblynZYhI

[2]: https://www.youtube.com/watch?v=NHEaYbDWyQE


But if the videos are longer then people stay on the site for longer! It must mean that they like it more.

Or, you know, your platform makes me watch ten minutes to find something that would take 30 seconds otherwise, so fuck your platform.


Same reason why I am absolutely terrified of playing any instrumental music on Spotify. I let one track run till the end and my next 2-3 discover weeklies will be filled only by instrumental music.


i used to instinctively turn on private mode every time i opened the spotify app (on desktop at least, since it was easy enough to get at)

with youtube i usually open a video in a incognito tab anytime im watching something random, otherwise it takes weeks of clicking "not interested" just to get rid of some recommendations. sad times


i basically ignore all the recommendations ¯\_(ツ)_/¯


I have the same experience with vocal music on YouTube Music, funnily enough. I greatly prefer instrumental music but you listen to one vocal album and suddenly all they recommend is singing.


You can use this behavior to your advantage though. I have my browsers set up to delete cookies when they are closed and from the surface this works for Youtube recommendations. If I open the site I only get very generic popular/pushed content.

So after I watch that instrumental music video, there will be a mass recommendation for other instrumental music. And that's nice, because I apparently was in a mood for that. And the next day, I open my browser again and they are all gone! I can dive into the music for the mood of that moment right away.


Huh... I guess I never really noticed, but you're right. I used to listen to entire genres on YouTube and let the recommendation engine pick the next video. I discovered some cool songs that way.

Now, maybe I'll type in a song title, listen to the song, and the next video is a live performance of the same song... Followed by a lyric video of the same song.


YT seems to have made some change to dump regular trending stuff into recommendations.

I really miss the option to tell it that I just really don’t want to see sports.


It's interesting that these companies go into these different product directions, while i assume they're both looking at similar metrics, and optimizing for the same outcome: engagement and number of users.

Somehow YouTube is seeing more engagement by showing more of the same, and TikTok is seeing even more engagement from showing fresh content.

Maybe the risk of an upward trend in outcomes, that blurs the fact that you could see even better trends by done things differently.


For me, YouTube is literally the same - I get the same videos thrown into my feed in repeat. Apparently this converts well for YT and the result is that my feed is maybe 10% genuine discovery. The rest is shit.


YouTube also does this weird thing for me where, for example, I watch a couple of videos of an android related YouTube channel about new phones, and then it’ll recommend me videos from 2008 about phones being released then. Lol


this is Hill climbing, right? https://en.wikipedia.org/wiki/Hill_climbing


Yes. How do you deal with hill climbing in product management? How do you know that a 10% improvement in outcomes is bad, and a different approach could have given you 50% or 100%? More experimentation, more 'how might we'? Google is known for trying different approaches (many shades of blue for a link), but somehow none of their experiments indicated fresh content is important? Or is it just a matter of product management 'playing it safe' at Google, where they know 10% outcome improvement is good enough to keep their job?


My guess is limited time frames within some standard A/B testing protocol. Give people very similar content over a 1-2 week period, and they'll watch more of it. Give people very similar content over a 6 month period, and they'll get bored and leave. If your testing protocol doesn't look for long term effects, you'll never see the longer effect in any of your tests.


Reddit made a change recently (in the last week or so) along these lines, to start injecting new subreddits in to your feed. They've picked up on this too.


Reddit is getting / is pretty good in my opinion


I've been a long-term subscriber to Google Play Music, now YouTube Music, and the change introduced and reinforced what you say. Each time I open the app it's to meet the same recommendations and automatic playlists; barely anything new. About time I got serious about transferring to Spotify.


Focusing on content rather than a social network.

Most videos you see on tiktok are probably not people you are following, while most on Instagram probably are

They test content on a small circle of users, if metrics are good they keep expanding

https://www.wired.com/story/tiktok-finally-explains-for-you-...


More data beats better algorithms, and Tiktok gets a lot of signals about what you might like based on how long you watch, how many times you watch, what you like, what you comment on, if you go to the profile page of a video you just watched, if you share, etc. Multiply that by all the videos you can see in a single hour, and Tiktok learns more about you than YouTube has in the past 5 years of use.


Absolutely not - I was using youtube for 10 years and every year my feed gets worse. Now the original content creators are dissapearing from my newsfeed and uts fetting invaded eith short clips of family guy. Its also will show me same clip again and again. It shows clips from creators i disliked. New uploads of creators i follow sometimes arent on my feed - its a shitshow. Half of comments are trolls and bots.


Here to say my experience is the exact opposite. My YouTube experience and recommendations have consistently improved over the last decade, and I find so much wonderful, informative stuff there.

TikTok, on the other hand, is a giant waste of my time. I have yet to find a single video I've liked from that platform; very little useful information there.


> family guy

Holy crap literally this happened to me few days ago


I made the mistake of watching a couple, what will happen next :)


I interviewed with them last year, and literally everyone I spoke to was brilliant. I think a lot of their magic is in hiring great people and having a very flat org chart. They get out of people’s way, and let them do cool shit.


Nothing makes me more depressed than thinking about the amount of human ingenuity dedicated to making teens spend as much time as possible consuming "digital crack".


Definitely don't think about all the ingenuity that's wasted on making you click or buy something then.


Potato potato.


Are they an American or Chinese company?


Chinese ! So maybe less focused on absolute numbers in terms of profit and able to be more focused on the product ? Even saying that feels so wrong, right, like US companies really cant enjoy building anything anymore ?


Chinese, but they're building recommendations teams in Mountain View and Seattle.


You’ve to remember that they have an (ancient == valuable ) source of customer data. They started off as a news aggregator and must have had a decade of customer data to utilise while showing these videos. I remember reading in the book (the attention factory) how fanatical the leadership is towards investing in ML talent


> I always wonder what Bytedance is doing differently compared with other social recommendation tools (aka why does no one rave about Twitter's suggested follows).

The thing is, the engine creates the content. Not quite as efficiently as a human producer saying "I want a 30 second video about X", but through the feedback loop of what gets promoted.

Twitter is run by libertarians who prioritise free speech, and therefore what Twitter produces is more and more intense political argument. Once the Tahrir square revolution and the color revolutions happened, that's what the place is locked into.

Instagram created the "influencer": young hyper conventionally attractive women photographed in beautiful locations, advertising products like energy drinks on their own account rather than a traditional modelling agency structure.

Youtube created the "let's play" and the "thirty to 120 minute political rabbithole" and the "shocked face thumbnail" genres.

Tiktok's secret weapon - and I'm not familiar with how it works in detail, but the effects are very clear - is a video editing tool, a content creation system, that is accessible for ordinary people. The evidence of this is that much of the "reels" content on Instagram's Tiktok clone has Tiktok watermarks on it.


I watched my nephew create a training montage for a toy. He takes in some video sequences, adds some music and gets something far more compelling and does so quickly. We just don't have a good natural understanding about what turns raw content into a good video. Tools like this help massively with that.


> Twitter is run by libertarians who prioritise free speech...

Oh wow, I wasn't expecting such a big laugh this early in the morning... thank you!


I haven't tried it, but my understanding from reading the experiences that it is about content not the creators.

Twitter, Youtube, Instagram, Facebook all about creators and social "connections" or follows to these. So you get the good with all the ugly and horrible. Want insights, art or jokes. Get the horrible political opinions and signalling too... And then see all the fights.


The massive data vacuum is probably part of it.


But that’s not at all unique to TikTok. Facebook, Instagram, YouTube, Reddit, etc. have massive amounts of data to play with as well. What else does TikTok have that these other platforms do not?


From what I've seen from Chinese tech companies, I've always noticed they tend to go one step even further in ignoring privacy. You think FB is bad, wait until you realize what Baidu, Tencent, etc do without even blinking. Probably because what the Chinese government demands of them, the only ones that stay standing are the ones willing to do that. And if your willing to do that, what is doing something similar when your not forced to?


Tictok have agreed to a $92 million settlement for lack of notification to users about data collection. https://www.nbcnews.com/tech/tech-news/tiktok-owner-agrees-9...


Perhaps they added a way to measure relationships between content, and expand the recommendation to several degrees of separation between related content.


No other social network lets users send a “dislike signal.


I think the important distinction in 'do I really want this' becomes whether I chose to do that activity, or if I kind of just got sucker punched into it.

I think there is nothing wrong with saying 'I am going to spend X time on social media and get entertainment out of it'. It is very different when you go to do one specific thing, and spend the next 2 hours in a complete daze in rabbit holes.


About a year and a half ago, I acknowledged how negatively some reflexive social media habits impacted my life and managed to extricate myself from them. A year or two before that realization I installed TikTok to see what the fuss was about, checked out the stream for a few minutes, looked up at the clock to realize it had actually been about two hours, and immeditealy uninstalled the app.

Astonishingly good attention and interest manipulation. Terrifyingly good, in fact. I don't moralize about getting lost in trash entertainment, but there's a significant swath of the population for which this format combined with these design patterns is a big problem.


Somewhere in the 80s some concerned parents in Jordache jeans just got kicked out of their mall's little town square facade where they were trying to hand out political pamphlets. And they went, "Oh, so that is the thing."

So... what's the thing here with TikTok? I mean it cannot just be that it's a personally-tailored time suck. What's missing? Is it pro-unionization rabbit holes?

I'm going to go ahead and guess "pro-unionization rabbit holes." If I'm wrong then the evidence will be me following the link someone gives me of a 2-hour TikTok unionization rabbit hole. I can't lose, internet!

edit: clarification


Your comment is incomprehensible.


I'll take another stab at it.

OP's critique seems to be that TikTok caused engagement with their preferred content for a longer duration than desired. That's not an interesting distinction from what happened back in the day with people going down Wikipedia rabbit holes. "Digital crack" is hyperbole if all it means is spending longer on the topics one already likes.

My question: is the TikTok algo consistently steering users toward/away from categories of content they would have otherwise had an interest in consuming? E.g., are there pro-unionization users who accidentally spend 2 hours watching pro-unionization TikToks? Or is the recommendation engine steering those users to 2 hours of some less contentious category?


You're talking to a transformer language model.


Really? How do you know? I had a look at their other posts and it doesn't seem like it to me.


Are you referring to some in-joke that I am not getting?




Consider applying for YC's Spring batch! Applications are open till Feb 11.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: