Start with a simple, interpretable model before reaching for deep learning

Rule #4 says to keep the first model simple and get the infrastructure right, and Rule #14 notes that starting with an interpretable model makes debugging easier. A logistic regression or a gradient boosted tree on tabular data trains in seconds, shows which features matter and exposes data bugs quickly.

The simple model then becomes the second baseline: a neural network or a large ensemble has to beat it by enough to justify the extra cost of training, serving and explaining it. Rule #40 adds a related warning: keep ensembles simple.