Answer these out loud before you look back.
One. Who makes a label, and what does that cost?
Two. Why can you not train a language model the way you train a cat and dog classifier?
Three. What does a database do when the record is missing, and what does a model do?
Four. Somebody says the model "knows" the capital of a country. What is wrong with the word knows?
Five. You are given a dataset of ten thousand images, all photographed in the same room. What have you actually got?