Real question: do you know how to describe a picture in a way that would be useful to humans but not as much to an AI? You can be sure that all those nicely labeled pictures are used to train AI models, and I would like to avoid diversion of a disabled-oriented feature for AI benefits if possible.
Yeah I know you can't escape the AI training, even this very thread will probably end up in a dataset sometime, but I kinda feel wrong doing a good job of describing my pictures for people, and other folks taking advantage of that for their profit. I don't work for them, it's not intended for that.
Not adding alt text doesn’t help you in this context. LLMs are already really good at imagine identification, what they need is more training data to *refine the images of those things they that gen AI models create.* So even posting an imagine without alt text gives them that.