1. Everyone makes a website.
2. Those websites submit articles to content indexes.
3. Part of your website hosts your own reader client:
3.a The reader subscribes to a set of indexes.
3.b It can also subscribe to individual feeds.
3.c Your client decides how to mix the feeds into displays, eg, you could have several indexes feed into one "feed" and have all of your family feeds in a second "feed".
4. Most of this is actually done via cloud hosting of managed packages, rented.
There's still many strong factors in favor of centralization, but if it's built around interop, then having a few large providers than mesh together is still an improvement and begins a competition to provide better services.
It's what happened with blogs and RSS anyway.
... aaaaaand, there it goes; the spammers already ruined it ;)
Like they did with email. Then you have companies come with robust AI tech and filter the spam, and also give good recommendations. A few of those companies come to dominate the market
... aaaaaand, there it goes; we're back to the situation we have now with Facebook and Google.
The situation for social networks would be strictly better if you could easily migrate accounts between FB, Google, etc and subscribe to feeds from multiple networks.
The situation you're calling out is exactly the point.
That's also my point.
1. You connect to friends and you either manually assign or "grow" a bunch of weights to your amount of trust in them and your interest in various topics they might post.
2. You and your friends make posts and assign "topic" metadata to the posts. The value of the system depends on your friends properly assigning metadata to posts. But it makes it easier to filter out extended families' political shitposting (based on trust and negging untagged posts).
3. Your feed is selected based on the weight of your friendship times the weight you base on the interest of the topic. If your best friend posts something about computers and you have positively weighed that topic it would bubble up.
4. Friends-of-friends can transitively receive a proportion of the trust score you have in your friend. Highly trusted friends will make you trust their friends more.
5. Down-votes and upvotes would train your local agent. They could also be fed back to the originator IF votes were signed by trusted encryption keys to prevent your friend from faking his vote count on the "post".
This decentralizes the network out of FB and democratizes the algorithms / system for what and how you want to receive your information.