Agreed, I click through them once in a while when I'm bored and it consists mostly of stuff I've already watched mixed with stuff I have chosen not to watch from channels I follow combined with random crap. The most annoying thing is that it just fills your feed with stuff similar to whatever you recently did. This is made worse by the fact that they only show a handful of recommendations. The only way to get new recommendations is to either wait hours/days for the stupid stuff to go away or to engage with their fiddly "this recommendation was BS" system for getting rid of recommendations. Google news has the same "more of the same shit" masquerading as machine learning style algorithms.
On a positive note, I like what e.g. Spotify is doing in some places and I think Youtube can learn from that. They provide some great content discovery mechanisms. Curated playlists, users that listen to artist X also listen to artist Y, these are the top tracks for this artist (based on play statistics), etc. Very useful. Very low tech. Very easy to implement.
If only they'd stop confusing geo location with my interests. E.g. their release radar is full of German crap because I happen to live in Germany (I'm actually Dutch). Youtube does the same "you are in Germany! Here's some random German crap!" style promoted content on their front page. Most of it is very obviously not even remotely close to anything I watch. Location seems to drown out recommendation signals and all forms of common sense. It is used to push locally promoted content regardless of the not so subtle hints that I don't ever engage with that (my German sucks; why would I?).
There are hundreds of millions of migrants world wide, many of them have spending power and they are ignored and under served by most big tech companies (Amazon, Netflix, Spotify, Apple, Google, etc.). This is surprising because they actually employ a lot of these migrants: they should know better. Getting content in the right language vs. the local language, localizing promoted content to the local language instead of what the user actually speaks and consumes on their service, etc. It's not hard and it's not because the content isn't there but because the algorithms are optimizing for the local lowest common denominator and most of these companies shoot.