How AI Works
Loop: collect data → pick a representation (features or pixels) → train or write rules → test on held-out examples → deploy → watch errors → update. Training means adjusting weights so predictions match labels (supervised) or find structure (unsupervised).
Board: 100 ham / spam emails. Split 80 train / 20 test. Learn. Score the 20. If test is weak, you overfit. That loop is “how AI works” for a fresher — not a 50-layer diagram on day one.
How AI Works — output — train has 4 labels, test has ['spam']. You score only on test.
data → train / test split
│
▼
model
│
▼
predict → monitorThe loop + why a test set exists.