Train a small model, then test its limits.
Choose a model task.
Label examples and train.
Test new cases and explain errors.
Activity
Train a tiny transparent classifier with preloaded classroom examples, test it on unseen data, inspect mistakes, and learn why people still need to review model outputs.