Expect this is a problem pattern that will be seen a lot with LLMs. Do I look at whether the data format is easily output by my target LLM? Or do I just valid

Expect this is a problem pattern that will be seen a lot with LLMs. Do I look at whether the data format is easily output by my target LLM? Or do I just validate clamp/discard non-conforming output? Always using the latter seems pretty inefficient.