What AI gets wrong

Hallucinations, bias, sycophancy, and how to catch them.

What it is

AI tools make up plausible facts and citations, reflect biases in their training data, and tend to agree with you. None of this is rare; it's built into how they work.

Why it matters

The most important AI skill is knowing when not to trust the output.

Key ideas

• Hallucinations are most likely on specifics: names, numbers, citations, quotes.
• Models tend to agree with the premise of your question.
• Ask for sources, then open them.
• Tell it it's wrong when it's right, and see if it caves.

Try it: Run a hallucination hunt

1. Ask for five academic sources on a niche topic you know well.
2. Look each one up in USC Libraries.
3. Record which exist, which are distorted and which are invented.