At what level of formal methods verification does the argument against AI-generated code fall apart? My expectation is that the answer is "never".

The subtext is pretty obvious, I think: that standards, on message boards, are being set for LLM-generated code that are ludicrously higher than would be set for people-generated code.