Stop 5 / 8 · ~25 min

World 3: Checking AI Answers

Confidence is not the same as correctness.

Goal: Learn not to trust AI blindly.

Lesson

AI can hallucinate. That means it can give an answer that sounds correct but is wrong, unsupported, outdated, or made up.

NotebookLM reduces this risk because it answers from sources you provide, but you still need to check important answers.

Use this verification checklist:

  1. What claim is being made?
  2. Is there a trusted source for it?
  3. Are names, numbers, and dates correct?
  4. Would a mistake cause harm, confusion, or cost?
  5. Should a human expert review it?

Example

AI answer:

This course guarantees certification in one day.

Problem:

That is a strong claim. It needs a trusted source. If the source does not say this, do not trust it.

Better wording:

This course is designed to help beginners prepare for AI fluency material. Completion time depends on the learner.

Practice

Label this claim:

Claude is always correct if you write a good prompt.

Choose:

  • Trusted
  • Needs checking
  • Wrong

Model Answer

Wrong. A good prompt can improve the answer, but it does not make Claude always correct.

Checkpoint

You are ready for World 4 when you can explain hallucination and use the verification checklist.