There's no inherent difference. It's people mistakenly generalizing after hating on current image and text generation AI models. And there's obviously a discussion to be made about how these models were trained unethically by stealing data, but a) that is irrelevant to their usefulness as a technology b) it was recently shown that you can train them with purely public domain data with good results (yogthos posted an article from Stanford the other day).

GenAI is not (inherently) a grift, unless you believe the following applications are nothing but grifts:

• machine translation
• image upscaling/restoration/coloring
• support chatbots
• etc