12 min · up to 60 XP

Spotting hallucinations

The tell-tale signs of confident nonsense.

Since the McKinley fiasco, Priya reads AI drafts the way a birdwatcher scans a hedgerow. Hallucinations aren't random noise: they come in species, each with its own field marks. Learn the marks, and suspect sentences start standing out before you've consciously read them.

Module 1 covered why fabrications happen. This lesson is pure fieldwork: three realistic drafts, each hiding one specimen. No theory first. You already have the theory. Binoculars up.

Spot the bug

Priya pasted her campaign report into the chat and asked for a client update email. Her report contains no percentages: she wrote "clicks noticeably up". One line below is a hallucination. Find it.

Click the line you think is wrong.

text

11. Hi Marcus, a quick update on the spring campaign ahead of Thursday's review call.
22. Open rates are up on last month, and both new subject lines are outperforming the old ones.
33. Click-through in the loyalty segment has risen 23.7% since we moved sends to Tuesday mornings.
44. Happy to walk through the detail on the call. Shout if Thursday no longer works.

One down. Priya's field guide lists five species. The over-precise number: decimals where the input had none. The unverifiable source: a named study or firm that thirty seconds of searching can't find. The almost-right date: real events, wrong year. The confident niche claim: bold assertions about your corner of the world, where thin training data meets your expert eye. And the deadliest: the seamless blend: one invention riding in a paragraph of truths.

Next specimen: one claim, pinned up for dissection.

This lesson is part of the full course

The free lessons are just the opening chapter. Unlock the rest of AI at Work to keep going.

Every AI at Work lesson, as it ships
New interactive lessons added monthly
Founding-member pricing, locked in
Learn at your own pace, forever

or keep browsing the free lessons