What should you consider before upload a revision to your arXiv paper when you spot an error?

Some journal policies can prevent people from uploading new versions of a paper based on feedback from their review process, at least for a certain time period. Some journals in some fields will even consider an arxiv version a "prior publication" of the paper, and as such won't bother to even review it.

Even if there's only a time delay, it's entirely possible that the author(s) simply forgot, as after 6+ months they are probably swamped in new projects, new semesters, etc. This can be compounded if the corrected version uses journal specific styles, and so would have to be edited to be suitable for the Arxiv.

Individual authors also use the Arxiv differently. Some of them use it simply as a way to announce their results once they've advanced far enough for a preprint. Such people are unlikely to bother with updating, especially if the paper gets accepted by the first journal it is submitted to. Others go further and view it as a paywall free way to maintain an accurate and up-to-date record of their work. These people are likely to update with corrected and published versions whenever possible, and may view doing so as an important (or even necessary) service to the research community. Many sit in between: they may update when significant changes in the results have been made, but won't bother with grammar and typo fixes alone.

In short: it's a matter of personal taste and preference. You don't want to submit an update for each typo you find, but you should at some point have a sense of "meaningful amounts of change has occurred, and I do not think I'll be making any more changes in the near future", at which point it would be appropriate to update if so desired.

Another reason is that the arXiv itself somewhat discourages overly frequent updates. For instance, after the 5th version(?), all future updates are no longer publicized in the mailings, so you might end up with the first few trivial corrections seeing more publicity than the future ones. I am myself no stranger to the "I'll update the arXiv once I've changed something significant" behavior, even if the preprint is years old and it is questionable whether any significant changes are to come.