The book builds up in stages: what machine learning is, using simple networks; then the efficient market hypothesis and what it means for anyone hoping to predict prices; then statistical inefficiencies, algorithmic trading and risk management. All of it with code and real series.
What I took from it
The lesson that travels outside finance: a test on past data nearly always looks good, and that tells you nothing. If you tune until the historical result looks handsome, what you achieved was a description of the past. I apply that suspicion to any promise built on history: demand forecasts, late-payment models, savings calculations. The right question is not how often it was right on old data, it is what it would have done on the day conditions changed.
The second is the gap between calling the direction and making money, which the book shows brutally: a model right more than half the time can still lose, because the wins are small and the losses large. Translated into a project: an improvement in an indicator is not an improvement in the result, and you have to show the bridge between them.
And a third, about data: chronological order governs. Training on information that did not exist at the time is the quietest mistake I know, and it happens constantly with ERP tables whose dates get updated in place.
What I left out
Nearly all the financial content, which is not my field and on which I will not opine. Also the speculative final stretch about a future run by artificial intelligence in finance: the most enthusiastic part and the least supported. And it is worth remembering the book is from 2020, with everything that implies about its technology.