the arxiv is good. treating every preprint you find, even one that is a preprint for an eventually published paper, as having the same rigor as peer-reviewed empirical studies is silly. mostly because there do end up being important and critical errors in arxiv submissions that for various reasons are not ever corrected except for the final journal version. arxiv papers should be relied on for something critically important without verifying. when you can, it's good to steal and horde the best, finalized version like sci-hub.

Arxiv is for preprints, maybe they should be retracted or not but generally whats up there is from before they're reviewed. My paper is on there before we got published - that's right, I'm an author of a math paper~

Probably they should still archive papers for sure aside from one place. That's part of how we protect this valuable data

It would be good to have the pre-prints on arXiv labelled retracted if they actually are (I don't know for sure if they do or not), but it is for pre-prints and something anyone can upload to even if it hasn't been accepted by a journal yet. So it can be used responsibly as long as the person looking at the papers knows that there is no quality control, and people can upload anything. Since, usually only academics really go there much, and its original purpose was a loophole to get around paywalling of journals. And that same loophole can get uncensored versions of these articles out there for people to read.

So, while it does have this opening for abuse, since mostly academics use it I wouldn't say that it is bad at the moment. So far it hasn't been a significant channel for much misinformation since most misinformation is through Youtube, Facebook, etc. posts rather than people citing bad pre-prints. And I am even saying this as a physicist which, maybe it is just the filter of my own experience, but I think is the field of science that has the most cranks publishing their own theories to arXiv, such as "the new theory of super-gravity that shows dark matter isn't real" or "why quantum mechanics proves conscious beings have souls".

me when I have absolutely no understanding of what an archive as a concept is

data rots

Hard to say, but "in the next 6 weeks" is not out of the question, and "in the next 18 months" a distinct possibility even if AI crashes and burns. There's wonderful rumors coming out of China's electron beam lithography teams on multiple fronts, most notably that one of the labs may have a practical achromatic lens.