LLMs have proven to be quite good at evaluating their own output. Think of things like "Think step by step" and how that improves output.