https://www.astralcodexten.com/p/how-did-you-do-on-the-ai-art-turing I did well, I got 30/42 or 43 I think? (~70%) Average was 60%
- Talk about which of your heuristics worked, which didn’t, and which will soon be obsolete. People point out that AI includes meaningless detail, but the more basic point is that detail is effortless for AI, so you should expect to see more of it (what else is effortless for AI?)
- Dall-E house style is more about this detail, and ig being overly rlhf’d into golden hour lighting. Apparently people think AI art is yellow now too (distinct from golden hour, maybe something about organicness?)
- Also AI doesn’t find the same things significant, or really know what’s good or bad/what people will focus on to quite the same extent. So there are lots of boring faces with beautiful backgrounds (since both are judged relative to the human artist face/background baselines)
See response: https://scottsumner.substack.com/p/school-of-rembrandt-fake-vermeers
Maybe compare what was so interesting about the amateur flash photography I sent to G?
Oh duh this is the same stuff as ornamentation becoming cheaper to produce, so the elites moved away from it. (if that’s actually true historically). This means we should see a similar move towards minimalist art being high status once again, as the plebs are able to make more ornamental kitsch. Probably the same applies across mediums. Long novels should also be slightly lower status, but I’m less confident about that. Certainly rhyming poems will be even lower status. The simpler supply increase price drops, there’s more ornate art art should go through without the complex status stuff
Oh! Fun parallel! AI art models build surfaces easily but struggle with physicals, they generate 2d but are still getting better at 3d physical coherence. This is clearest in video models. Anyway point is they are very fluent at surfaces, but under the surfaces there’s nothing there. In the same way outward symbols of mattering (details in part of a painting or ornamentation) that used to be costly are now cheap so we care less about them. Point is just that surfaces/outward signs got cheap which detached them from their original source of value (either simply scarcity or a causal connection to a human caring enough to make it (which aren’t unconnected of course))
I feel like this has a place but idk where aesthetic responsibility I guess. The thing you make when you make art with AI is a strange object. Take cicada. Maybe half the bi grams, similar amount of trigrams were spat out by R1. But I picked the prompt, evaluation prompt, sorting algorithm, basic idea, and selected the lines. So who knows what “I” made. This contrasts with a clay sculpture. You can ignore the distant causal chains, gaf, just note that the immediate function is you personally take not-art (clay) and make art. But lots of complex industrial or post industrial art is like this. Take a video game even a single dev one, what’s the art? The source code? What is it without the compiler, or the console that runs it. Are we all holding to the auteur theory when we dismiss AI art?
- This is not the same concern as it’s ripping off artists during training, but the obsession with provenance does remind me of a Sotheby’s auctioneer
R1 poetry is not quite tradition and the individual talent or blooms misreading but something like it. The traditionnis not mine but it comes when I call.
People say we will have individual movies, but people like to share media or feel they’re watching the same thing. You would need bot generated comments. The fact people want individual algorithms doesn’t mean they want individual movies.
If AI is about to get better at personalised editing of existing works, creating higher highs, not consistently excellent works, is going to be the important thing