2026-09-20 08:06 UTC
there's not many things that get me going so fast as hearing that LLMs are useful as accessibility tools.
they aren't. where they somehow work, they are at best adequate. and it's never done with accessibility as a main goal, usually just something half made-up to make the detractors shut up.
now imagine what we could have, if a promille of a promille, one-hundred-thousandth of the money put into the autoconfabulators was spent on fucking actually developing accessibility tools.
Replies (1)
-
@maccruiskeen@social.linux.pizza 2026-09-20 12:26
@mawhrin@circumstances.run This may be slightly tangential to your complaint, but I oversee ebook production for a uni press. Our conversion vendors are all-in on developing AI-based systems for providing alt-text. We've tried it for some of our backlist (where we can't go back to the author), and as you might guess, it's not great. So I have to review and edit the results. It will not surprise you that when it is wrong it can be very wrong. I've got a list of a few hundred titles I need to process. This will not be fun. I envision that at some point, the reading systems are going to try to implement doing this on the fly. That's going to suck.