Using LLMs to find errors and using LLMs to produce error-free artifacts are two extremely different levels of challenge.