The selling point behind TikTok is that by using the app normally (i.e. watching what you want) TikTok can figure out what stuff you want to watch.
But pretty much every suggestion that comes out on HN for people who complain about the algorithm on sites like this is to do unnatural things to "train" the algorithm: be very careful about how long you watch videos, click on videos that you think are relevant even if you don't want to watch them, watch things you're not interested in to prevent falling into a rut, avoid click on things you're curious about if you don't think you want to watch 100s of them to avoid tainting your recommendations, etc.
If that's "working" I think I'd agree that TikTok was mostly hype.
I spent some time disliking every 5-second video being recommended to me, and the problem went away. It was easy.
I'm another one who's tried TikTok on a number of occasions and it's always failed badly with me, worse than any other site I can think of. It never improved despite people telling me it eventually would.
My impression is that it was making too many assumptions about me based on the wrong signals or something.
But to your point: wouldn't there be a survival bias of sorts with assessing recommendation performance? I didn't like it and left, so presumably the people who remained liked it.
It's obviously popular so there's that, but it seems circular to say it's working for the people who like it and stay. You're losing information about opportunity costs of lost users.