Hacker News new | ask | show | jobs
by exe34 4 hours ago
Nowadays I imagine OCR and vllms would solve that? Tesseract is incredibly fast and accurate.
1 comments

Accessibility is not just about getting text out. It's about the navigation flow, integration into various OS features, Aria attributes in HTML and what they do. Plus you may not count this as "accessibility" but things like integrating fully with multi-lingual text entry methods.

It's solvable, except for what isn't solvable because the browser doesn't expose it, which can be solved by fixing that too. But it's a lot more work than meets the eye. And building the layout engine is hard enough in the first place. Give me a week and an AI budget and I'll produce you some sort of layout engine that works when fed exactly the sorts of inputs I anticipated, but to build something that survives contact with the real world is going to be well beyond something you just prompt your way around today.