The crux of this seems to be that "reviewing code written by other people" isn't the same as "reviewing code written by LLMs". The "human" element of human-written code allows you to utilize social knowledge as well as technical, and that can even be built up over time when reviewing the same persons' code. Maybe there's some equivalent of this that people can develop when dealing with LLM code, but I don't think many people have it now (if it even does exist), and I don't even know what it would look like.