7 min · +80 XP
Why It Lies (and How to Catch It)
Hallucination, demystified — and your three-second defense.
By the end of this lesson
You'll trust AI correctly: a lot, but never blindly.
Sometimes AI states something false with total confidence. This is called hallucination, and once you understand why it happens, it stops scaring you.
The core fact: it predicts plausible next text. Usually plausible and true overlap. Sometimes they diverge — and it will happily generate something that sounds perfect and is wrong, because 'sounds right' is its actual target.
Verify anything with consequences: facts, numbers, money/legal/medical claims, code that touches real data. Ask it to double-check itself or show its work. Trust it like a sharp intern, not an oracle.
In Claude Code the confident-but-wrong moment looks like: “Done — built and tested! ✓” …and the thing is broken. The defense is identical: your eyes on the result. ALWAYS open what it built and click around yourself. Its 'it works' is a claim, not a verdict — the verdict is yours.
Mindset locked
AI is a force multiplier you supervise — not an authority you obey. Hold that and you'll outperform people who either fear it or worship it.
Author
Evgeny Arsentyev
PhD · Chief Product Officer at a healthtech company