What a Trained Model Is: Weights, Not Rules · seed 1 · A4, ink-friendly. The answer key prints on its own page for grown-ups.

A model is numbers, not rules

Computing · Artificial Intelligence · ages 14-15
Name ______________________   Date ____________

  1. What sits inside the tiny flu model?

    • A shelf of medical books
    • Three weights plus a bias
    • A list of typed rules

  2. How does a written rule differ from a trained model?

    • The rule uses more numbers
    • The rule is typed logic; the model is fitted numbers
    • They are exactly equal

  3. What stays fixed while training runs?

    • Every single weight
    • The training examples
    • The multiply-and-add recipe

  4. A spam filter sorts well but cannot show why it moved one mail. Why?

    • Spam has no patterns
    • No readable lines exist behind the answer
    • Inboxes hide the rules

  5. A patient is misclassified. What does training do?

    • Deletes the patient record
    • Rewrites the recipe from scratch
    • Nudges the weights and bias toward the right answer

  6. Training ends today. What happens to the numbers from here?

    • They keep drifting nightly
    • They reset to zero
    • They freeze until retraining

  7. More examples can change a model without anyone editing logic.

    Circle one:   True   False

  8. A vendor claims their model reasons through written rules you could audit line by line. What is wrong?

    • A trained model holds fitted numbers, not auditable logic lines
    • Models never make mistakes
    • Spam filters need no training
LightMySky · lightmysky.comW1-mt_3_1toZUxak-s1

Answer key

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

A model is numbers, not rules W1-mt_3_1toZUxak-s1

  1. Three weights plus a bias · Its whole brain is three weights plus one bias number.
  2. The rule is typed logic; the model is fitted numbers · Rules are typed by people, while models store numbers fitted to data.
  3. The multiply-and-add recipe · Training moves the numbers, never the recipe itself.
  4. No readable lines exist behind the answer · No logic lines were written, so none can be shown.
  5. Nudges the weights and bias toward the right answer · Each mistake nudges the numbers a little toward the right answer.
  6. They freeze until retraining · Once training ends, even the numbers freeze.
  7. True · New examples nudge the numbers, while no logic is edited.
  8. A trained model holds fitted numbers, not auditable logic lines · Nobody wrote logic lines, so no audit of lines is possible.
Worksheet · LightMySky