Heads up: this browser isn't saving anything for this site, so progress and settings reset when you leave the page. Everything still works while you're here.
Some of NeuralQuest’s cast are still learning. When you teach them, YOU learn it best — catch what they get wrong, then quiz them.
Teach Veer
You be the teacher — Veer is still learning. Helping Veer figure it out is the practice. (Veer's progress stays on this device; nothing is sent anywhere.)
🎓 In a minute you'll be the teacher. Get ready to teach Veer: generalization vs. overfitting.
A model that's perfect on its training examples may just be memorising them. The real test is new data it has never seen.
You've helped them one at a time. The real understanding is how the ideas CONNECT — teach that next.
Tag (labeling — every label is a human choice and a meaning-making act) ×Skew (bias and data fairness — whose data is in here, whose is missing, who decided)
A model learns to sort photos, but the labels were all written by one small group. How do Tag and Skew connect?
Tag then says: “The way data is labeled is always neutral and never affects a model's fairness.” — is that right?
The link: labeling is a human choice, and that is one place bias enters.
Who labels, and how, shapes what is fair.
Drill (training loops — once, again, again; and knowing when to stop) ×Veer (generalization vs overfit — trained here, tested here, now go somewhere new; does it still know the way?)
A model scores perfectly on its practice data but fails on new data. How do Drill and Veer explain what went wrong?
Drill then says: “A model that scores perfectly on its training data is guaranteed to do well on new data.” — is that right?
The link: testing on new ground tells you when to stop drilling.
Train, but stop before it just memorizes.
Capstone — build one fair model together
Three steps toward a fair, working model. Bring in the cast member for each.
Label the data thoughtfully. Who?
Who do you bring in?
Check it works on new examples, not just memorized ones. Who?
Who do you bring in?
Weigh whether it should be built at all. Who?
Who do you bring in?
The synthesis: label with care, prove it generalizes, and weigh whether to build it — a model that is fair and honest.
Label, generalize, and weigh — a model worth trusting.
