I see it as basically Cunningham's Law. It's easier to see the LLM's attempt a solution and how it's wrong than to write a perfectly correct solution first time.