As I write this, the question already has one vote to close (but no comment as to why). So, I'll start by stating that I think concerns about arxiv are reasonable for research mathematicians, and that you're likely to get more thoughtful answers on MO vs academia.SE, because many other fields do not have the same culture of communicating via preprints that math has. Furthermore, we have had several MO threads about arxiv that are still open, e.g., 1, 2, 3, 4, 5, 6, 7, 8. I will interpret the question to be about staying on top of the volume of stuff on arxiv, concerns about correctness (since the OP mentions this), and general early-career advice regarding arxiv. In terms of "tips for reading" I'll say that reading an arxiv paper is just like reading any other math paper. I assume the OP already knows how to read math papers, e.g., the importance of the introduction, finding relevant results, checking the proof, keeping track of notation, etc.
Here are some of my personal considerations regarding arXiv:
I try to check the arXiv in my area at least once a week to stay up to date on what people are doing in my area. I read the title of every paper, read the abstract of those that interest me, and download the most interesting to read fully when I've got time. At the very least, I read the introductions of those most relevant papers. About 25 papers remain visible on the "recent" page, e.g., see this. If it's right in my wheelhouse, I try to read the preprint fully the following weekend, so as not to fall badly behind.
If an arxiv paper overlaps with work I'm currently doing, I might email the author to discuss, or to invite them to the local seminar in my area.
You can also subscribe to arxiv email bulletins, that include when a new version of a paper is posted.
For any arxiv paper that I think I'm likely to cite, I keep my eyes open for when the journal version appears, and download it, because sometimes the theorem numbers in the arxiv version do not match the published version.
I hate to admit this, but I do pay attention to who is the author of the paper. If it's someone I've never heard of, making big claims, I treat the preprint with more skepticism. Before basing any of my future work on it, I make sure to read it carefully and check if it seems correct. For papers by authors who I know well, I usually give them the benefit of the doubt, and don't check things carefully until I really need the result. I basically attach the new knowledge onto my existing knowledge as something that so-and-so proved, and hope I remember that the preprint exists when I next ponder that particular branch of my memory tree.
It is wise for junior researchers to come up with a file organization system that works for them. Mine is to have a giant folder called "papers" where every paper is saved like "White, Yau, 2017, arxiv v1, Smith Ideals of Operadic Algebras in Monoidal Model Categories." That way, if I can remember the author names or keywords in the title, then I can probably find the paper using the mac spotlight feature or going through the folder in alphabetic order. This also means if subsequent versions come out, they can be saved near the first version. I imagine others have FAR superior organization than I do. An old friend used to use Mendeley, and it could attach keyword tags, etc.
Sometimes a paper has been on arxiv for a very long time, has a well-known author, and has tons of citations. Here is an example off the top of my head. Sometimes this happens with an author who has died or left math. Having been around a long time, with lots of citations, I think the community has essentially verified all the claims, and the paper can be cited with the same level of certainty as if it had been peer reviewed.
Whenever I get a paper back from a referee, I update my bibliography to be sure I cite published versions of any arxiv preprints I had previously been citing.
If you are just starting off as a research mathematician, I advise you to NOT post anything to arxiv without your PhD advisor's explicit approval. Why? Well, for one thing, first impressions matter a lot. If your preprint has errors, or fails to cite some prominent people, readers might form a negative impression of you. For another thing, you don't necessarily know what kinds of things will be viewed as impressive or worthwhile by the community. Lastly, you might not know the culture in your subfield regarding junior people getting scooped, and you might not know how easy/hard it would be for a reader to read your preprint, understand your methods, and push on to the next step of your program that you and your advisor were hoping to do. I think it's good to target putting stuff on arxiv at the same moment that it's ready to submit to a journal, and when you're certain there is little danger of getting scooped (e.g., you've already worked out the first two or three papers in the series).
Having preprints on arxiv the last year of your PhD might help you when you're on the job market, to show that you really have produced good research.
If a new version is posted to arxiv and you want to know what the author changed, consider using a PDF comparison tool like this one.
If I notice another researcher citing the arxiv version of a paper of mine that was subsequently published, I usually email them with the publication citation so they can cite the published version instead.
When one of my papers gets accepted, I try to post the last version to arxiv, so the only difference between it and the published version is the use of the journal style file. Relatedly, it's good to update the arxiv journal ref to point to the actual published version so people who see the paper on arxiv realized it did actually get published (plus when and where).
When I see a paper that's very interesting to me, by a very new researcher, I sometimes write them an email to encourage them in that line of work. Like "Wow, this is so exciting, I tried to prove this myself once and couldn't. Wonderful work!" I think math research can be a lonely profession and the math community might be a bit biased towards negative feedback (like, "I knew all this 10 years ago" or "What's new here? This seems like it all follows in a straight-forward manner from X.") and I'd love to shift the culture to be more encouraging. I especially try to encourage people doing the kind of work that I'd love to see more of. Maybe that will help the author feel less alone in the pursuit of this knowledge, and know that others are interested and rooting for them.
Anyone uploading papers to arxiv should know that their tex code is public. Therefore, it is wise to remove any comments to yourself that you don't want others to see. It's easy to find lines that start with %, and strip them out.
I have to dash off now but I might add more if I think of any.
