Hacker News Reader

Comment by blisseyGo | Hacker News Reader

Slightly related but some Redditor "created a NSFW TikTok clone, with just portait-mode porn gifs, some with sound. (works best on mobile)":

https://www.reddit.com/r/SideProject/comments/enn048/i_creat...

View on HN

These are neat, but in my mind, TikTok is far more about the ForYou algorithm than the fact that it's a scrolling feed of videos. The latter is fairly easy to do, as showcased by this very thread. Any decent engineer can throw together a demo like this fairly easily. TikTok's huge success though seems to be due to how good it is at predicting what people will like, which is something that even Youtube struggles to sometimes. Of course predicting 10s videos is much easier than predicting 5m videos.

The algo has been failing pretty bad; "hide videos from this user" does not work anymore, i suspect it can only deal with a limited number of ignores. Last summer I did an experiment, tapping "not interested" on any food-related video (I was on a diet) and yet i kept being shown food videos.

The former issue may be intentional. I've seen many recommendation algos not quite letting users shoot themselves in the foot. It's like how if you subscribe to 1000 channels on Youtube, it'll start only showing you the videos you actually watch instead of all the videos.

The latter is strange, but generally for me it has worked pretty well, without even needing to rely too much on "not interested". For example, in March I started playing Animal Crossing, and a week later my feed was 50% ACNH content. Then I stopped playing around May, and a few weeks later, I was quickly down to maybe 1 a day and now I'm at 1 a week maybe.

That being said, I like and share videos frequently, so that may be a good signal of what I'm currently interested in to the app.

Reply on news.ycombinator.com