Day 07: Why confident AI answers still need evidence

The dangerous hallucinations are not the strange ones. Those get caught. The dangerous ones are fluent, confident, and almost right, a real policy with one wrong number, a real process with one invent