Well, the tiktokers like algorithms. The simplicity of a single page that can instantly respond when you find a new interest, for example you go down a rabbit hole about sourdough bread and it starts showing you that content.

This kind of an algorithm is undeniably better for discovery compared to custom feeds.

An algorithm like this needs to run on a centralized server. You need to analyze all content with AI to label things thoroughly so that you can properly recommend relevant content.

And as much as nostriches hate this, a mainstream app needs to have censorship. It should be voluntary, but a mainstream app should hide pornography, gore, death threats, encouraging self harm, etc.

If I could build my dream TikTok clone right now, I would build a client using a caching server like primal, and use the caching server to consolidate content from popular relays to label the content using AI. I would have it create user interest profiles based on each npub’s interactions, with heavier weighting for recent interests. Every time the user refreshes the home page, the server would generate a list of events that match the users interests, showing them content they haven’t viewed, and serve that list of events to the client.

It would have 3 levels of censorship:

1. Default: no porn, or death/gore

2. Strict: same as default + slurs and “hate speech” hidden

3. Disabled: Nostr experience, fully uncensored (CSAM would not be relayed by the caching server)

It would have the ability to connect to any relay. Relays that aren’t already read by the caching server would be connected to directly, but they would not be visible in the algorithmic feed, only in the chronological feed.

In the case of a caching server outage, the client would read directly from relays, without the AI labeled algorithm. Maybe a lightweight algorithm could run locally in the client in the case of a caching server outage.

I can barely code a webpage at this point so I really don’t know what I’m talking about 🤣