> Our heuristics for judging human output are useless with LLMs.

We used to call the negative side of these heuristics "code smells", but I see no one has used that term yet in these comments. Code smells are what the post is referring to, what LLMs get rid of.