@zersiax At the very least, it can give normal keyboard-driven instructions instead of "click here, click there". As for accessible code, I guess yes, at least in my experience it helped in this too. But Florian is also right, it needs to be trained on code with better quality.

@menelion @zersiax yeah, I'm not sure that until we build a more accessible web (and start caring for reducing errors, some because of legal reasons others out of passion) that what it's trained on will improve. I would imagine seeing some errors going down in things like WebAim Million would help AI as well in certain areas or parts of knowledge more than others, but that's a super slow trend and someone would need to observe it.

@Tamasg @menelion ehh custom prompts can really help, too. Like... a big reason why devs don't use semantic HTML and native controls is because they can't figure out how to style those exactly like the design briefs tell them to style it, I've made that point before. If AI can be made to essentially spit out CSS resets that are always appicable in all browsers (this is a pipedream but hear me out), at that point we can essentially go back to HTML = structure and CSS = presentation but really the better way to fix that is for browsers to agree on how one can strip away all the browser-generated styling and roll your own, or for companies to not want to stand out by making their controls look different from the norm (again, this is a pipe dream) xd

@zersiax @menelion yeah, and most design schools don't train about that dev language to tell them what semantic-type control is best for the job, either. I think designers will always want that flexibility (as do some developers,) and accessibility can erode it (talked about this in engineering.atspotify.com/2023…) - AI's job will be to balance this in better ways but will need humans.