Comment by shakna

I don't recall where, but I've heard that before in the past. Perhaps in the kind of slop that makes the rounds on LinkedIn.

There is sort of a good reason for it, in the past. Before the overhaul, Microsoft Speech used to skip Facebook posts, and read the alt text instead. It is now, however, more sane. Facebook was pretty darn bad at accessibility in its early days. A lot of intermingled broken spans for text, causing stuttering and other issues.

However, today, most reading systems prefer the "title" attribute, to the "alt" one. If title exists, it'll read that and skip alt. Some always skip alt, regardless of it exists or not.

Figure and figcaption are about the only way to get good and consistent behaviour, but you don't really control how those happen on most social media platforms. You throw everything you can at the wall, and see what happens. And it might change tomorrow.

Today, I'd say the above is bad advice. An image description is a good practice. Repeating yourself, isn't.

Visually impaired people are more than willing to tell everyone what they need as far as accessible UIs, myself included. Barely anyone listens. Let’s not let LLMs be the next thing that people that don’t understand the actual problem try shoving in as a solution.

Wheelchair users / people with mobility impairments rightfully scoff at the myriad “concepts” you see now and again of mech suits, wheelchairs that can climb stairs, etc. “Just give us a ramp! This is a solved problem. Your alternative is just sci-fi fetishism!” Still, it keeps happening. LLMs are increasingly becoming the same thing for people with visual impairments.

I don't need the text of the page. Thats easy, and I already have it.

But information has a hierarchy, usually visual, and that hierarchy needs to be reflected. LLMs are famously bad at structure, especially any tree with significant depth. RAG is not enough - hallucinations become common at depth.

My response now, to you, is in a semi-structured node graph. I know a reply has happened, because of the dangling children. I know who made it, and what they said, by cell attributes in the spans, surrounding it.

Don't worry - AI is being shoved down accessibility's throat, like everywhere else. FSCompanion for JAWS, NVDA has an OpenAI plugin, and VoiceOver has it builtin.

Why do I hate it? Because when it _doesn't work_, you can't tell. You don't know if it is hallucinating data, and cannot verify the response. If it is the mode of communication, it is all you have, making every failure a catastrophic failure.