Comment by astrobe_

Amusingly, LLMs are good at detecting patterns because they are modelled after the human brain, which is very good at detecting patterns (this e.g. is annoying when you are a game developer who wants to reuse assets). People are still able to tell if the source of an artwork (or text) because LLMs tend to output "patterned" text (like the famous "load bearing"). That being said, the false positive/negative rate of humans for this task has yet to be measured.