The secret sauce of TikTok’s recommendations
shaped.ai
shaped.ai
The Facebook feed used to work like that before 2013 something. You could clearly see the difference when they started optimizing for ads or what ever metrics they used. Posts could not go "viral" anymore in the way it used to. Zuckerberg capped the vitality so that companies had to pay I guess.
So, this obviously is a poor filter, which does not keep me engaged and which proves that Facebook is NOT a social network as to be social, it needs to act as such, and I never hear from my closest people over there anymore!
How hard it is to see that a post is from my mother-in-law and multiply its weight by 100, or one from my sister and multiply by 1,000?
Another thing that Facebook never puts any effort into is photos of children under 13! They had a feature called "scrapbooks" where you can create profiles of your children with the promise of those turning into profiles after kids turn 13. Well, they've abandoned this feature and people kept tagging children as their parents! Not sure how this served their facial recognition, but it inconvenienced us parents!
Eventually they had to pull those features out. Privacy concerns which lead to people sharing less data? Not sure.
Even now a lot of people get surprised when I put their full name into search box and show them what photos are available via simple search - they completely forget or don't know even that when they are tagged their photos are searchable by their name. I mean now most people I know don't share much anymore, facebook is not that cool anymore.
[1] great demo of graph search https://www.youtube.com/watch?v=DySRn7c736c
"TikTok’s Secret ‘Heating’ Button Can Make Anyone Go Viral" https://www.forbes.com/sites/emilybaker-white/2023/01/20/tik...
They also don't just optimize for consumption, but also for new users. Don't have concrete proof, but it seems they help new users get more views to give them that taste of "15 minutes of fame" as early as possible.
How many people have created and published videos on YouTube and got maybe a few tens or a couple hundred views over years.
TikTok new users gets a couple THOUSAND views and the user is like "people like my content I MUST make more!!" And now they're hooked trying to get the same high view count as before.
I would assume (as I have never used the platform) it's also a way for TikTok to grab you financial information "you earned x$ from you first video, please supply you financial information and a scan of your ID for payment.
It's an addiction.
I agree that it's likely intended to give new users that hit though.
"You _could have_ earned x$ from your first video. Good job! You're on your way to the Partner Program please supply your financial information and a scan of your ID for future payment."
Surely the temptation to do some for profit is too big.
It would not surprise me if TikTok goes the same way as Facebook on this when they get comfy, to milk som short term dollars.
Why do we let kids use this again?
I don't, no matter how much they ask. The most I've acquiesced to is a small amount of YouTube time per day. It gets harder the older they get. My daughter can now just use her school laptop (while at school, at least) to watch as much YouTube as she can find time for. I'm not real happy about it, but part of raising kids is learning to accept the steady transfer of their autonomy over to them. I just keep gently harping on how low quality this kind of content is, what the motivations behind it are, etc.
It's about curating this yourself for your child (as opposed to giving them unrestricted access), and then slowly giving them control and hope they have good taste :)
Same reason we eat junk food. Its easy!
Also as another potential source of proof, tiktok is really effective at pushing alt pages to people's FYP. Which is interesting because their old account will stagnate, but their alts will get pushed quite aggressively despite the content being very similar. So obviously it's not a content issue, tiktok is evaluating them equally and thinking you'll like it, but their old account(s) don't show up as much anymore.
Tiktok isn't as overtly aggressive with this method though, it seems like there's a cooldown period of sorts where it might start pushing you out again if you remain committed to the platform. Still, it's severe enough for people to make note of it and feel seriously demoralized by it.
There might, of course, still be additional new-user weight. But I think this is at least a part of it.
Mouth cancer is no joke, I knew a dude without a lower jaw because of that
also turned into a few skin cancer issues in there.
don't smoke and wear sunscreen, people
But very relatable and I too am baffled by the appeal of these shorts.
As a recovered TikTok addict, I can supply that it was the jokes for me. There was a lot of content that got delivered to me that hit my sense of humor just right. I’m sure you could track increased oxytocin in active users just like with narcotics.
But at the very least, maybe it was just trash compared to the experience of not watching a short video on your phone while riding on an escalator. Trash compared to just looking around. Would staring at your shoes, or the head of person in front of you, or the environment surrounding you have been a better experience than those short videos.
It's TV 2.0, but caters to ADHD types or people with borderline ADHD. If you use The Internet in any extreme way, you already have an ADHD brain. It's the only logical way to handle information overload and deal with the vastness of the web IMHO. ADHD as a term is normalized because of The Internet. It's not a scary thing to have anymore. It's the new normal.
Atleast she is using good posture for phone use. Give her that.
I also noticed the quality of the filters. They really make ordinary people look quite a lot better than they do in real life.
Young girls have feeds full of eating-disorder content, and young boys have feeds mixed with misogyny like Tate. Not all teenagers watch either, but these recommendations are a disease plaguing the platform.
Imagine a world where there are thousands of companies out there that produce these strange little unpleasant pellets of food that they want people to eat. They don't want people to buy them—the company is happy to send out pellets for free because they make money when the pellets get eaten. (Let's not worry about how that could make economic sense. Maybe the pellets are made of sequestered carbon and the company is selling carbon credits.)
So all these companies want to get millions of people eating their pellets, but people don't want to. They don't taste very good and they've got, like, better shit to do with their lives.
But people do like eating other food. And usually they have to pay for it, which kind of sucks.
An opportunity exists here, and "social eating" companies pop up. These companies will ship you food for free, and you can eat as much as you want. What joy! The only catch is they've mixed some of these not-very-tasty pellets in with the food they send you. The social eating companies get paid by the pellet companies, which is how they're able to make and send you food for free.
Here is the interesting question: What kind of food do the social eating companies make? You might think that since they want to get as many people eating their food mixed-with-pellets as possible, they would make the best food they can. And, indeed, they do want to make food that is scrumptious, compelling, and mouth-watering. But what they don't want it to be is satisfying. Because once you're sated, you stop eating. That's the last thing they want because the more pellets you choke down, the more they get paid.
So what they make is junk food like chips and cookies. Each serving is a single tiny delicious bite, but as soon as it goes down, you're even hungrier than before. Its high in anticipation and desire, but low is satisfaction and satiety. You crave it, but once you have it, you don't actually feel any better. If anything, the craving is even more intense.
Now apply that metaphor to information. The most junk-food-like content is the stuff that triggers anxiety and anticipation if you don't watch it: fear of missing out, question-inducing "You won't believe...", alarming "You're doing ___ wrong...", etc. The content doesn't make you feel good if you do watch it, it makes you feel bad if you don't. Because that way, when you do watch it, the nagging anxiety goes away a bit, but you still don't feel "done" and still want more.
that which inflames or causes controversy gets sent around or reposted.
clicks == money.
as does "sponsored content" aka "post this to everyone who meets demographic groups X and Y". just so happens that some of those people want folks angry about some topic.
Most of the videos seem to be 1.5 minutes long.
These systems show paid content mixed with user-generated content, promoting harmful advertisements, political propaganda, or any other ideology anyone with enough resources and desire to influence large segments of society is able to take advantage of.
Let's not forget the Cambridge Analytica scandal, and not doubt that there are _many_ more such companies operating without public knowledge. This is all done legally, and within the terms of service.
This is even without mentioning the rampant data siphoning and shady multi-billion dollar data markets that exist in the background, that generate most of their revenue. After all, all this technology and optimization is serving this end goal.
These platforms are not just harmful to individuals, but to society as a whole. I hope one day governments catch up to the harms they're doing, and regulate them just as they did for Big Tobacco and many other truly evil industries.
But how does that happen when our government officials (at least in North America) seem technologically illiterate, and rely on Silicon Valley for their re-elections and donations (therefore becoming more susceptible to lobbying)?
- Jeff Hammerbacher
A video game programmer might sat.. "all of the best minds are working on silly games" Why are so many bright minds wasting trying to get us into space when we have problems that need to be solved here.
The best minds solving problems is only part of any solution anyways. You need the best salesmen, best leader and right moment to bring in change that mat make lives better.
Framing advertising as "people trying to share beautiful things with you" is some hell of a newspeak.
Don't turn it into "we just followed commands" or "we were tricked", own whatever you're working on.
What decisions are on these guys? How to name some variable? How to receive mole salary? Or what manager's/marketologists' decision not to code?
Let grass roots content creators actually distribute their content. Let the best content go to the top. Give everyone a chance.
Facebooks algorithm:
Support influencers, corporations, and whatever pumps the ad machine.
Whether that behavior is healthy or not... I'll leave that up to people more educated than I. My guess is that trading the user's attention span for short-term engagement isn't worth it, though.
Ie. tell ML to trick someone to spend 14 hours watching youtube every day, and for some small percentage of users, it will actually succeed!
For those people, it's as addictive as drugs. They spent all day on youtube rather than going to work, going to school, caring for their kids, eating or even sleeping! Can you imagine the size of lawsuits that would be heading youtubes way when those people realise they've effectively been enslaved by an algorithm??
Leadership of the big companies put an end to that, instead trying to focus on other metrics, and trying to get more users to each spend some time on the platform.
Well it seems TikTok didn't get the memo...
Mine is full of cats, food, attractive women, and also ADHD content. None of those are remotely surprising to me, except that there's not more ADHD content (though I suppose the feed implicitly is).
My facebook feed is a ghost town. Almost zero people i know posting, a couple of straggling groups that remained relevant, and just bs ad driven nonsense.
Facebook and Youtube on mobile are so packed with unappealing scams that I stop scrolling after 5min
I used to fill my free time with world of warcraft spending more time in Azeroth than i did with friends or at school or at work. But i came to the realisation it was just escapism from a boring country/region and a waste of my time. But this is also made me realise how lucky i was to be born in such a boring country/region.
I would like to see more about how they are formulating the problem, I know there is work lately in "sequential recommendation" that is focused on generating a sequence rather than scoring content items, I'd like to learn more about that.
If you had to pick one number for judging classifiers that is the one.
"Accuracy" is a really bad one because it is not meaningful w/o careful thought about the problem. For instance if you have a diagnostic system for a disease that has a .01 prevalence, you get 99% accuracy if you say nobody has a disease.
Practically a person might want to operate a classifier with different thresholds depending on how bothered they are by false positives vs false negatives, the ROC shows what choices are have available with a particular and the AUC of the ROC gives you one number that characterizes the quality of the classifier overall. A system with an AUC under say 0.55 hasn't really learned anything at all, at 0.6 it is showing signs of life, looking at those curves I'd say Tik Tok is beating me solidly. In this case it is limited by the problem being fuzzy (something I thumbs up today might get a thumbs down tomorrow), something up in the 0.95 level could be attained for something like "is this article about the outcome of a sports game?"
Right. If your AUC is in the 90's you usually either have a bug, or a predictive task that isn't very challenging.
For some other like “is this an astrophysics or materials science paper?”, getting 0.95 or better was straightforward in 2005.
But in the 1 year I've interacted with the app I've already seen the effects of the content creators learning how to game the algorithm and get presented with some contents that were made in a format only to get engagement, similar to a clickbait article.
The result? It has spoiled the whole experience for me and I uninstalled the app.
And this is how those platforms slowly crumble to their own weight as the crowdsourcing becomes a business for creators and, little by little, content becomes more and more superficial, fake, commercial and uninteresting for the audience, who migrates to another app and everything starts again...
TikTok is optimizing for user generated content that connects deeply with their users. Thats what they reward money which is largely derived from TikTok's "gift economy", and recommendations and therefore views are largely based on the gifts. This is in stark contrast with something like YouTube, which is optimizing for 'time spent on YouTube' which drives their creators to make longer clickbait type videos, regardless of how dissatisfying they are.
The recommendation algorithm described here could be tuned to serve any 'objective algorithm', like time spent, profit to Bytedance, or whatever- but because of TikTok's choices it is making a really fun an enjoyable experience for users.
Why would people spend time watching dissatisfying videos?
> content that connects deeply
I'd say "parasocially".
But how does the app measure that? It makes decisions based on a few seconds of watch time. It could optimize for "show videos that trigger donations", but that's kind of trivial.
- https://news.ycombinator.com/item?id=33494796
Related paper:
- https://arxiv.org/abs/2209.07663
Related GitHub:
What's addicting about it? Admittedly it kept me entertained the first day or two that I used it, but from that point forward it never wavered from the content that entertained me that first day and soon it became really boring to see the same type of thing over and over, leaving no remaining appeal.
Because everyone talks about how great the recommendation engine is, and me wanting to find great content, I went back to it after a while to see if they fixed the problem. But no, still the same content that bored me away from the app the first time.
https://www.pewresearch.org/internet/2022/08/10/teens-social...
I’ve spent several hours using it and there seems to be very little of interest to me (a 53 year old Canadian living in Texas). It did show me a video of a guy doing some great welding (stack of dimes!) and I watched the whole thing. The algorithm decided that I must want to see every welding video ever uploaded and I really don’t. One or two is enough.
I concluded that TikTok isn’t for my demographic, maybe by design.
Maybe TikTok appeals more to teenagers who have an underdeveloped sense of self. I remember at that age conflating popularity with accuracy and authority. Now when I watch TikTok it's just an exercise in frustration that nothing is cited and it's all about presentation over substance. And that's when the algorithm isn't showing me strange videos in a desperate attempt to keep my attention, a la spiderman/elsa YouTube mashups of yore.
ByteDance runs TikTok, but they also sell recommendations as a service. The paper that this blog post is based on explicitly talks about that service in the abstract, and never even mentions TikTok.
Now, if I were to advertise a service that contains some of the ingredients of my other, very popular product, I would surely mention that fact ... but this is not the case here.
regardless, if this described how drug dealers engineer the distribution of their product to optimise addiction, we would not be celebrating it as an engineering marvel.