There were some articles re how generative ML doesn't understand composition (it doesn't understand anything, but it's good at pretending for common stuff) that

There were some articles re how generative ML doesn't understand composition (it doesn't understand anything, but it's good at pretending for common stuff) that came out when the vision models were popular last year, e.g: https://garymarcus.substack.com/p/did-googleai-just-snooker-one-of https://garymarcus.substack.com/p/did-googleai-just-snooker-...