Comment by Gander5739

If the output can't be trusted, and you use another llm whose output can't be trusted to check the untrusted output of the first llm, then you're back where you started.