Why a Chatbot Can Be Confidently Wrong · seed 1 · A4, ink-friendly. The answer key prints on its own page for grown-ups.

Confident words are not proof

Computing · Artificial Intelligence · ages 12-13
Name ______________________   Date ____________

  1. A chatbot gives a book title, author and year you never heard of. What should you suspect?

    • An invented reference in a correct tone
    • A book that is simply too new
    • A typing slip in one letter

  2. What is the difference between a likely sentence and a true one?

    • There is no difference at all
    • Likely wording can still be false
    • True sentences always sound unlikely

  3. A very confident tone means the answer is probably right.

    Circle one:   True   False

  4. Why does the model invent facts instead of staying silent?

    • It wants to trick the reader
    • It copies a rival chatbot
    • It picks likely words with no sense of truth

  5. An answer reads well but one fact in it looks odd. What would have caught it?

    • Reading it a third time
    • Checking that fact against an outside source
    • Asking the model to repeat it louder

  6. You checked one claim and it failed. What next?

    • Trust the rest even more
    • Check the rest the same way
    • Delete the whole answer unseen

  7. A calm detailed paragraph names a study with full details, yet no library lists it. What happened?

    • The model invented a real-sounding reference
    • The library index is always incomplete
    • The study is too recent for any shelf

  8. Two students get the same answer. One checks a date in it against an encyclopedia, one asks the model if it is sure. Who checked better?

    • The one who asked the model, since it knows itself
    • Both checked equally well
    • The one who used the encyclopedia, since it sits outside the model
LightMySky · lightmysky.comW1-mt_TMHNkprrks-s1

Answer key

For grown-ups. Fold this page away before handing over the rest.

Confident words are not proof W1-mt_TMHNkprrks-s1

  1. An invented reference in a correct tone · Invented names and titles arrive sounding fully real.
  2. Likely wording can still be false · Sounding likely and being true are separate matters.
  3. False · Confidence carries no information about truth.
  4. It picks likely words with no sense of truth · Likely-word picking has no truth sense built in.
  5. Checking that fact against an outside source · Only a check outside the model can catch invented facts.
  6. Check the rest the same way · One failed claim means the rest need the same check.
  7. The model invented a real-sounding reference · Full details in a calm tone still allow pure invention.
  8. The one who used the encyclopedia, since it sits outside the model · A check must come from outside the model to mean anything.
Worksheet · LightMySky