@Stefan Bohacek @BeAware :veriweed: @ᴚ uɐᗡ @Sampath Pāṇini ® I don't have mixed feelings. From my personal experience, and for my own purposes, it just plain sucks. But that's to be expected. It would require extensive pop culture knowledge plus absolutely extreme niche omniscience in my special field. Like, if I've built something in-world, an AI would have to immediately know everything about it.

I've actually had LLaVA describe an image I've described first. Yes, it was an in-world rendering.

In comparison with my own 25,271 characters (no, I'm not kidding, go check the link), LLaVA's 558-character description was vague, it was painfully lacking and incomplete, it didn't explain anything because it had no idea what it was really dealing with, and it was even glaringly wrong in some points. It would not helped anyone understand the image.

I've got my doubts that ChatGPT can do much better than LLaVA. And I've got huge doubts that ChatGPT can produce something more detailed, more informative, more explanatory and more accurate than me from any in-world image I'd post. After all, ChatGPT would only have the picture while I could look around in the virtual world itself and see things that ChatGPT can't see.

#Long #LongPost #CWLong #CWLongPost #FediMeta #FediverseMeta #CWFediMeta #CWFediverseMeta #AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta #AI #LLaVA #ChatGPT