Alpaydin's Machine Learning gave me a clean way into models, data, training, testing, and generalization without turning the entrance into a ritual sacrifice to notation.
The part I enjoyed most was the patient glossary-like quality of the book. Terms are introduced with examples and plain English, which is exactly what you want when a field is full of words that people repeat before they understand them.
My edition is from 2016, which makes the timing funny now. It reads like a serious snapshot from just before generative AI started walking into every room with wet shoes and a confident smile.
The comparison between unlimited intelligence and unlimited energy stayed with me. It is not a decorative analogy. It points to scale, control, responsibility, and the uncomfortable fact that powerful systems do not become wise just because they become capable.
For me the book worked as grounding. It reminded me that intelligent systems are still systems: assumptions, data, objective functions, errors, tradeoffs, and engineering decisions before the demo starts looking like magic.