Raw LLM Responses

I feel like the direction people describe as 'better' for AI art is actually the direction that makes it completely worthless. I had quite a bit of fun early on with DaliMini because the ways it made mistakes (especially when my prompts were just absolute strings of nonsense) made these vague suggestive images that worked a bit like liminal spaces. You could watch the sun blurring into a guy and recognise a similarity so bizarrely expressed that you hadn't thought of it because the AI wasn't thinking, it was just smashing key words together with no idea what they meant. It was interesting to speculate about whether the sudden inclusion of wavy lines said something about the synonym you'd just substituted in and its relation to the entire corpus of the internet, or was just random chance. As AI art is sanded down into just copying people, the emergent possibilities of sorting a large corpus of images programatically vanishes. I don't need a bad imitation of art accessible online, I already have access to the originals. It's the same with the term 'AI' in the first place. At every point, idiots want to paper over the interesting material differences between human interactions with art and machine ones and pretend that they've invented life. It's depressing not just because other people buy it, but because it's totally sanding out any unique possibilities a large corpus study (which is essentially all generative AIs inputs are) could have for our noticing of possibilities (questions like 'What proportion of images of bedrooms are blue and does this vary by search language?' 'Is a skyscraper taller than a spire?' other weird specifics like that)