Overview
AI assistants sound confident even when they are wrong. This session looks under the hood at how LLMs retrieve and synthesize information, why hallucinations happen, and builds a practical fact-checking workflow. In lab you will draft research writing with AI support — and then catch its fabricated citations.
Learning objectives
- Explain how LLMs combine training data and live search to answer questions.
- Define hallucination and recognize the situations where it is most likely.
- Fact-check AI output using verification tools and primary sources.
- Use AI as a research assistant without letting it write fiction into your work.
Session agenda
- 6:00 – 6:10Warm-up
Dataset chart share-out.
- 6:10 – 7:15Lecture: How LLMs search, hallucination risks, fact-checking
Training data vs. live retrieval; why models fabricate; a live hallucination hunt; a five-minute fact-checking workflow.
- 7:15 – 7:25Break
- 7:25 – 9:00Lab: Trust but verify
Draft a short research brief with AI support; extract every factual claim and citation; verify them with search and citation-checking tools; score your AI's accuracy.
- 9:00 – 9:10Wrap-up
Worst hallucination contest; preview ethics and bias.
Materials
- Access to a search engine and one AI assistant with browsing.
- Library database access (campus login) for citation checking.
- Companion reading: Elements of AI, Chapter 3.
Homework
Pick a topic you know well and ask an AI assistant ten factual questions about it. Score its answers, and bring the most convincing wrong answer to class.
Detailed slides, lab worksheets, and demos for this session are in progress and will be posted here before class.