Assuming “correctness” as being a hard criterion will greatly limit your ability to get powerful and very useful output from LLMs.

A person doing some knowledge work will fail on this criterion. But they still are employable and generally regarded as being useful.