← Back to course
AI Explorers 3-5 / Week 07 / Explorer Quest
6/6
Week 07 · Test the Model

Explorer Quest

Explorer Quest: Family Test Lab
// Keep some cards back
⏱ about 15 min

Explorer Quest: Family Test Lab

Raven hides a button in one fist. "Which hand?" she asks. Rocket guesses left. Wrong!

"Let's make it a real test," Raven says. "What do you notice we should count?"

Nova blinks her lights. "Try ten," she says. "Then fix your trick and try ten more."

  1. A training set teaches a model. A test set checks it.
  2. Test cards must be new to the model.
  3. Accuracy is how many are right out of how many were tried.
  4. Errors are clues. Fix the model, then test again.
  5. Real AI is built by people, and it can be wrong.
◇ FAMILY GUESSING TEST
Pick a guessing game, like guessing which hand holds a button.
One person hides the button. The other guesses ten times.
Keep tallies: right and wrong.
Talk about a trick that might help, like watching for a clue.
Try ten more guesses. Compare the two scores.
RoundRightWrongGuesses
Round 110
Round 210
For a grown-up
Ask: did your score change in Round 2? Why might that be?
Ask: why do builders test a model on examples it has never seen?
Celebrate any mistakes as clues, not failures.
QUEST CHECK
  • Read the question.
  • Tap your answer.
You got 6 right out of 10 in Round 1. How many were wrong?
In Round 2 you got 8 right out of 10. Did you improve?
Why keep some cards hidden when testing a model?
What word means how many are right out of how many tried?
WHY THIS EXERCISEAccuracy tells us how well a model did on a test.
← Friday