Comment by layer8

That isn’t necessarily true. LLMs could be good at finding issues that human miss when coding, while being bad at catching issues that they themselves miss while generating code.