@davidrevoy @tiredbun @voxel I've seen enough blatantly wrong LLM-generated alt text and spell check results that I don't think the technology is fit for purpose. Any purpose. AI "hallucination" is not a solvable problem: It's the exact thing LLMs are designed to do. So, with all due respect to the person being quoted here, I don't trust anyone who would entrust something as important as accessibility to AI.

@davidrevoy I've seen a few artworks on fedi with AI-generated alt text and it pretty much reads just like this. Sensible sounding description, wrong about various details of varying importance. (I get not being able to *write* your own alt text, I guess, but you could at least read it to see if it matches what you drew.)