What is a common mistake when working with Large Language Models?

Treating model output as guaranteed truth without validation can introduce hallucinations, incorrect decisions, or unsafe automation. In practice, the value of this concept comes from understanding where it belongs in the architecture, what problem it solves, and what constraints or tradeoffs it introduces.