12 · Check what you learned

1. What is the core idea behind machine learning?

A programmer types in every rule by handThe computer finds patterns from lots of examplesThe computer already knows everything when it's builtIt's magic that can't really be explained

2. In ML terms, what is a "label"?

A clue the model looks at, like colorThe right-answer category you're teaching, like 'apple'A bug in the modelThe webcam button

3. "Garbage in, garbage out" means…

A faster computer always makes a smarter modelA model is only as good as the examples you train it onYou should delete your training data when doneModels work better with no examples

4. Using Teachable Machine, what do you need to start training an image model?

A paid account and a coding backgroundNothing but a browser — it's free, no code, and no account neededSpecial hardware you have to buyPermission from Google for each project

5. What's the honest way to test how good your model is?

Re-check it on the exact photos you trained it withTest it on new examples it has never seenTrust the highest confidence bar and skip testingOnly test it once, then never again

6. A model scores 100% on its training photos but only 40% on new ones. What's the problem?

UnderfittingOverfittingPerfect accuracyGarbage out

7. Why might a face-recognition model work worse for some groups of people?

The model dislikes them on purposeIts training data didn't include enough examples of them (bias)Faces are impossible for any modelIt ran out of battery

8. Which of these is a real-world example of a trained model — 'features in, label out'?

A spam filter sorting junk from real emailA light switch turning onA calculator adding two numbersA printed paper map

9. A model gives a confident guess. Confidence means…

It is definitely correctHow sure the model is — which is NOT the same as being rightThe model is brokenIt has finished training

10. Which is the most responsible, honest way to share a model you trained?

Say 'my AI is never wrong' to impress peopleBe honest about its accuracy and limits, and respect people's data and consentUse it to secretly decide who looks 'trustworthy'Train it on classmates' photos without asking