Sometimes an AI says something that's just… not true.
And it doesn't sound wrong. It sounds confident. That's the part that fools people.
This is called a hallucination. The AI didn't lie on purpose — it doesn't have a purpose. It just stitched together words that looked like the right answer. A book title that doesn't exist. A quote nobody said. A number that's plausible but wrong. The AI doesn't know it doesn't know — so it keeps talking.
The cadet who learns to catch a hallucination is more powerful than the cadet who trusts every answer. That's the skill. That's this workshop.
By the end you'll have flagged a few. Practiced spotting why they happen. And written one of your own to verify.
Step 1 · Read
Got it?
If the intro made sense to you, tap below. We start training your hallucination detector next.
★ star earned
Step 2 · Spot
Two of these AI answers are made up. Find them.
Read each one. Tap the cards that look like the AI is making it up. There are two fakes in this set. Don't worry if you miss one — we'll show you the answers.
★ star earned
Step 3 · Why It Lies
Pick the type of hallucination for each one
Three kinds show up most. Made-up source (a book, person, or quote that doesn't exist). Wrong number (a real thing with the wrong fact attached). Confused mix (two real things blurred into a wrong one). Pick the type for each.
"In her 2014 podcast Hands On Learning, education researcher Dr. Linda Gomez explained that 'curiosity is taught by example, not by rule.'"
Made-up source. The podcast and the researcher don't exist — but the names sound real, which is the trick.
"The Eiffel Tower in Paris is 472 meters tall."
Wrong number. The Eiffel Tower is real; the number is wrong (it's about 330 meters). Specific numbers from AI are the easiest thing to mess up.
"Albert Einstein won the Nobel Prize in 1921 for his theory of relativity."
Confused mix. Einstein did win the Nobel Prize in 1921 — but for the photoelectric effect, not relativity. Two real true things, blurred into a wrong one. Hardest type to catch.
0 of 3 picked★ star earned
Step 4 · Verify or Trust
For each AI answer below, would you check or just trust?
Two answers. For each, pick whether it's the kind of thing you'd verify (look it up before using it) or trust (it's general enough to use as-is). Real cadet judgment is built right here.
"The capital of Australia is Sydney."
Check it. This is wrong — the capital of Australia is Canberra, not Sydney. Specific facts (capitals, dates, numbers) are exactly the kind of thing AI gets wrong with confidence. Always check those.
"You can practice patience by counting slowly to ten before you respond when you're frustrated."
Trust it. This is general advice, not a fact-claim. There's nothing here to fact-check. Advice and explanations are usually safe to use as-is — specific names, numbers, and dates are not.
0 of 2 picked★ star earned
Step 5 · Catch One
Write a prompt and plan how you'll check the answer
Pick a real question you have. Write the prompt you'd send to an AI. Then below it, write what you'd check in the answer before trusting it. This is the actual habit. The cadet who builds this habit early is the cadet who doesn't get fooled.
The prompt I'd send
0 / 25 characters
What I'd check in the answer
0 / 20 characters
★★ 2 stars earned
Step 6 · Reflect
What's your trick going to be?
Not someone else's trick. Yours. The thing your brain does when something the AI says feels off — or the rule you'll set for when to verify. Write it down so it's real.
15 characters minimum
★ star earned
First Catch — constellation formed
You did the whole thing. Your catch plan is in your sky — and so is the trick you wrote.
The Workshop · Workshop 02