Hacker News Reader

Comment by falcolas | Hacker News Reader

What I'm trying to say: It's more complicated than the parent makes it sound. The parent makes it sound like "if you're on TikTok, this is what you see."

> TikTok livestreams are a bizarre experience that doesn't get fully conveyed by this article.

> It's a wild show that you don't ask for, it gets thrust on you every once in a while when scrolling.

However the reality is that what you see on TikTok really is a direct reflection on you. It's not an accusation, it's just an acknowledgement of the truth.

TikTok's algorithms are scarily-effective (dramatically more effective at tailoring than all of your other examples), and thus what you see is indeed a direct reflection of what you watch. Any single video (ads excepted), or even trend, just doesn't appear globally on TikTok.

View on HN

> However the reality is that what you see on TikTok really is a direct reflection on you. It's not an accusation, it's just an acknowledgement of the truth.

It's a direct reflection on what tiktok's algorithms think they know about you so far, modulo what they think they know about the contents of the videos they are showing. They have a good recommendation engine, sure, but it works on average over large populations through the limited funnel of video interactions, and their video understanding and inventory is similarly limited. This is even before considering that they clearly add some kind of extra exploratory weight to new content they don't have a lot of data about from you.

Their machine learning is basically fancy statistics on watch data, not a peering-into-your-soul Oracle, it's very possible it gets some people's preferences very wrong and is still profitable for larger population segments.

I find my recent experiences with the algo amusing. It very heavily weighs new creators that I may have watched a single video all the way through, so much so that perhaps every 4th video would be old content (couple of months old) from this new creator. It gets super annoying super fast that I have to resort to blocking these individual creators (marking as "show less" does absolutely nothing for me). I also note that the algo seems to run out of interesting content for me after about 20 minutes or so - at which point the videos are further and further away from my interests.

More on this point, there was a great article by the verge I think on how different tik tok was on the same city border ukranian vs russian.... wild war vs bliss

I feel like their ML is scary good.

Should maybe post on a throwaway but: it will show me the gay version of tit-tok, let's just call it putting a bunch of sweatpant season thirst traps in my feed.

Often videos with like 100 views or even less.

Their ML is likely categorizing d*ck prints, visible nipples, and a bunch of other human monkey brain dopamine drivers.

Instagram will recommends shirtless people. But TikTok takes it way farther. And probably a good reason it got so popular to begin with. People are creeping on kids.

I've seen several of the creepy / disturbing live videos described above - the begging family and the baby with the large deformed head specifically. Each have been shown multiple times, despite immediately reporting them. So this isn't a unique artefact of OP's experience. There clearly are ways of bringing these videos, which are clearly pretty far outside any normative preference pattern, to the front of the algorithm.

Shouldn't have to point this out - but I do not search for, like or watch anything remotely similar on Tiktok.

These videos are punishment for watching videos flagged as unapproved.

Think about what you watched right before these videos, its coded to being a feeling a disgust so you assioate that with the previous videos.

YouTube does the same thing with ads of kids dying in hospital, screaming for help on the start of 'controversial' content. Most of the time it makes you click back and watch something else.

That's a hell of a claim. Any citation for that? I could see Tiktok doing something that extreme, but it would be incredibly bad press for youtube. Simpler explanation might be that lower priced ads run against dubious content and that those tend towards the more spammy and extreme.

There is no way to completely avoid a certain type of content in an algorithmic feed. As good as the algorithm is, unless they have a "no variation or A/B testing" policy (doubtful), eventually you'll run into something that doesn't conform to your "profile."

> However the reality is that what you see on TikTok really is a direct reflection on you. It's not an accusation, it's just an acknowledgement of the truth.

I really do not think this is true, and is what the parent is getting at. Yes, the videos shown are a reflection of _past videos_ shown to you, and your reactions to them.

That does not mean that they are a _direct_ reflection on you, or an acknowledgement of the truth. The first videos shown to you, or a random stray video with off-content, whether you like them or not, can have a strong biasing effect.

TikTok’s algorithm is deeply flawed to draw any conclusion about. Its a deep reflection on what other people in your content branch accepted, not you.

Simply looking at the comments on a video you disagree with, just to see if others are appalled or brainwashed is something that the algorithm will interpret as deep interest in that kind of content. I have to go find the “not interested” button. Its sad because all those other people are really stuck in that rabbit hole.

> What are you trying to say? That the reason this person see this content is their fault?

You can control the feed by tapping on a clip you don't like and selecting 'Not interested'. Less successfully by immediately swiping to the next clip. In this way I have got rid of live streams, cats and girls doing dance or PoV trends.

But if you do in fact have a quick peak at the live streams, cats and girls doing dance or PoV trends, TikTok will keep showing them.

> However the reality is that what you see on TikTok really is a direct reflection on you. It's not an accusation, it's just an acknowledgement of the truth.

If this was the whole truth we could as easily excuse facial recognition algorithms failing non-white people by simply saying that non-white people are more difficult to identify.

The algorithm is a human-made thing and subject to conscious intents or subconscious biases of its makers.

Reply on news.ycombinator.com