I hate to suggest something so "in", but I do like the way StackExchange does things in the sidebar, a "you may also be interested in" presumably LLM-driven. Reddit as well, but IMO SE does it better.
Now training that requires a lot of user tracking data, that we don't have and shouldn't start collecting in one place.
Anyvworarounds come to mind?
something like that could be done in a dumber way by developing an algo that would show similar results by keywords and other stuff. then you could show some and get data on most clicked through links and show them as priority
We have the data on the relays and that isn't in one place, and we don't actually need to know anything about the user to track it. It's simply public data that we could analyze and mark, manually or with AI. We don't mark the user, in other words, we mark events. If the feed is not npub-focused, this is enough, as we would be suggesting threads, not persons.
We already have this with bookmarks and labels, and we shall be adding embeddings.
A user could actually curate his own AI-driven feed by marking events himself, and then asking the AI to give him "events with similar content to the stuff I have marked concerning topic Y". Since AI can scan the content field, it doesn't have to rely on hashtags. He could also include WoT markings or AI-markings.
