This is the lesson the hype skips.
Claude is sometimes confidently wrong. Treating its output as automatically correct will get you in trouble. The fix is a habit, not a feature.
Everything in this lesson is true today and may be less true in two years. But the habit it builds is the durable thing — it’ll outlast any particular model.
New word incoming. Look up “hallucination” → in the plain-language reference — along with the other AI terms you keep hearing. No jargon, no quiz.
How Claude is wrong
A few flavors, in rough order of frequency:
- Hallucinated details. It cites a file that doesn’t exist. It references a number that isn’t in your spreadsheet. It invents a quote. The shape of the answer is plausible; the specifics are made up.
- Stale knowledge. Its training has a cutoff. Recent events, new versions of things, a change you made to a file five minutes ago — it might miss those unless it actually looks.
- Confident wrong-ness. This is the dangerous one. It doesn’t always know when it’s guessing. You’ll get answers that sound certain but aren’t.
- Misreading scope. You asked it to change one thing; it also changed a related thing that looked related but wasn’t.
The good news: most of these are catchable in a couple of seconds, if you look.
The verify habit
Pick one thing to do every time Claude tells you something nontrivial:
- It says a file contains X? Ask it to show you the exact part. Or open the file in the app and look yourself.
- It made an edit? Read the change it shows you — the app puts the before-and-after side by side — before you accept it.
- It says it did something and it worked? Check the result yourself.
The habit isn’t paranoia. It’s the same instinct that makes you glance in the rearview mirror before changing lanes — cheap, automatic, occasionally lifesaving.
Reading is safer than writing
A useful frame for the rest of this course:
- Asking Claude to read something and explain it is low-risk. The worst case is a wrong explanation, which you spot when the next thing doesn’t make sense.
- Asking Claude to write or change something is higher-risk. A wrong change persists silently. Now you have a problem you didn’t create.
The next level is Read with Claude for this exact reason: build the habit on the safe side first. Then you start changing things.
What’s next
You know what Claude is, what it can see, what it asks before doing, and where it falls down. Time to put it all together on a real task.