Here's a video by a youtuber about all this: https://www.youtube.com/watch?v=bm-Jjvqu3U4
I believe the idea of moving towards a more AI/ML based recommendation system is interesting because it frees Youtube from the burden of defining what is worthy of being featured (and featuring everything that's relevant to each particular user). In a way it could democratize content discovery. But I don't see it working as well as it could be.
I have a feeling that Youtube may want to be able to promote certain youtube "stars" because, let's be real, they are the ones driving revenue. But I chose to believe a world where content discovery is more fair is possible.
I will say that this has less to do with the shitty way subscription works nowadays. The idea behind that is: if users want to follow certain creators, they should be able to. Moving that towards a AI/ML system seems just inefficient at achieving that goal. I don't know the rationale behind the decision here.
I had organized my subscriptions into a series of groups, which was a pretty hidden feature that they subsequently removed.
I have since replicated that by putting everything into a series of folders in the google replacement RSS reader, innoreader, I have been using.
YouTube sort of gets me to stay on the site by autoplaying vaguely similar videos and randomly inserting totally off topic videos that match my previous viewing history.
I feel like this scatterbrained approach to throwing videos at users does very little to foster go to the site out of the blue habituation, though it does make the site somewhat sticky once you are there.
I don't know if Medium can do anything similar to get the habituation working, maybe and editor's choice curation system, but that seems to fail compared to aggregators like HN or reddit. They could probably promote some stickiness by appending full stories right at the bottom of existing ones via infinite scrolling, if that even works better than just having the typical 'click here to see the top 10 secrets of X' junk at the bottom.