I hear people say that a lot, but I don't experience it myself on Amethyst. My "Follows" looks noticably different to "Global" - less bitcoin, less spam, more nature photos. I still have content on topics I don't care about, but they're by people who other times post content I am interested in.
My feed is also good, but most npubs have clients front-loaded with accounts and feeds. They didn't organically decide what to read.
And, if you ask for OtherTopics, your feed will be extremely thin, as they only really have data for Bitcoin Twitterati and stuff. I'm not on their map, nor are most of the people I follow. New people don't see most of the threads I'm seeing.
Makes sense. That's very much a client problem, and one that should be easy to fix using relay data. My "minitru" could classify topics quite well, I suppose instead of filtering out bad/reported content we could try filtering in wanted/interacted content.
There's seems to be a massive curation hurdle, tho, as measuring what most people are looking at, would just give you variations of Primal's trending feed. Most people see my selfies, as those trend for mysterious reasons, but they have no idea that I actually have any other topic and there's currently no effective way to signal that.
Interactions are the signal there, just as Kind-1984 was for filtering
If the client had an onboarding process involving a lot of swipe-left/swipe-right, would ppl do it? To train their minitru algorithm
I would so do that. 😂
We could have an option somewhere to re-train based on recent interactions, too
That would at least be very interesting as an experimental client, or an experimental option in a client.
What I'm not keen on is centralised categorisation of topics. Obviously clients are doing it now and poorly, but it cannot be done well, for many reasons, and is also very susceptible to misuse for ideological / personal reasons.