As an alt text user I will say that the sort of text you get from image OCR (i.e. what this thread is suggesting) is often lower quality than copy-pasting the text in a screenshot. In particular, text in columns or grainy scans often doesn't parse correctly. So this sort of tech isn't a panacea
Oh, I was dealing with crappy OCR back in the late 90's, I'm well aware. I do try to eyeball it first, and of course anything funky get edited in a notes app beforehand. Sidenote: I just used Google Lens on an old fuzzy image of a 1922 NYT clipping and I mostly just had to edit out hyphens.
Absolutely, if the original text is available it’s more accurate to copy/paste from that. Most folks are sharing an image they found elsewhere instead of an original screenshot, though, and this feature means there’s very little excuse to not make a bare minimum effort to add alt text when they do.