Day 5: Validation and Catching Hallucinations
Open the lab the instructor calls out from the slide. Auto-saves to this browser — refresh-safe.
Three labs today — and the week's wrap. Your instructor will tell you which one to open and when. Each one auto-saves to this browser — refresh-safe.
Catch the hallucination signature. Verify with source.
Three short AI outputs, each paired with its source. For every sentence, pick a verdict — clean, fabricated figure, fabricated citation, or confident generalization. Reveal the answer key when every sentence is marked. The trap is the generalization.
Cross-check against the source — twice.
Two tabs. First (5.2a): a financial summary with figures, side-by-side with the source ledger — mark each figure ✅ Found, ❌ Not found, or 🧮 Arithmetic wrong, and write where it lives. After lunch (5.2b): an AI contract review against the source LC — mark every claim ✅, ❌, or ⚠️ Paraphrased-changed. Auto-saves across the break.
Walk the four checks. Stake your name.
One pre-loaded scenario — a polished AI committee briefing + its source. Walk the four checks (figures, citations, generalizations, arithmetic) one at a time, log every finding, set your reject threshold, and answer the stake-your-name question. Print to take it home. Week 1 toolkit recap at the bottom.