Daniel Micay

An LLM can definitely be useful to blind users not only in a screen reader but as a camera-based app. LLMs are very unreliable and make up nonsense though. It's not something they should have to depend on. It would be a useful tool to work around things being inaccessible but they can be accessible.

The feature being possible doesn't mean that it will be available to most users. It would be useful to those users to have both the alt text with your perspective and the LLM where they can ask it questions about the images to get info you may not have included yourself. They're different things.

Screen readers often support OCR for getting text from images but not always. There will probably be screen readers integrating support for an LLM to describe images in the near future. That can't know the context, purpose and meaning of the image from your perspective, only mimicking training data.