Published 1 year before the second AI winter. This looks like a general-audience book, but most of the things you would find in an ML textbook published in 1986 would still be applicable today. The fundamentals haven't changed, just some new techniques and more computing power being available. Hinton popularized backprop (which is basically just an application of the chain rule from calculus) on neural networks in the 80s, and that's still how the errors are propagated during training of the newest LLMs.