[nextos]

I'd say Kevin Murphy's two-volume PML book is a wonderful outline of almost everything that concerns ML these days: https://probml.github.io/pml-book. Note there is a ton of companion code to recreate figures and discuss concepts. My other favorite, https://d2l.ai, is much simpler mostly aiming at modern CNNs and LLMs. It is really polished, and code is embedded within the text.

AI is very broad. I think the future is neurosymbolic, and these two books only cover a tiny part of symbolic, mostly concerned with probabilistic and causal models. See Murphy vol 2 sections V-VI. Lots of interesting ideas for symbolic AI can be found in the SAT and theorem proving literature.