Books I’ve been eyeing / trying to read include The Sequences, The Selfish Gene, and Superintelligence.
These works are written for a popular audience, they only teach you to talk the talk. I think it’s better to read textbooks and solve exercises, that way you learn to walk the walk. For some topics, like AI risk, there aren’t any textbooks with exercises; but you’ll still do good by learning adjacent topics like logic/computation/ML, for which there are good textbooks.
A good strategy is to read a chapter, then solve all exercises not marked “very hard” before moving to the next. Otherwise—no reading ahead. If some exercise is blocking you, you can peek at the answer, but only after spending 5 uninterrupted minutes trying to solve the exercise.
If you’re trying to learn probability theory, I think you’d indeed be better off with Jaynes’ Probability Theory: The Logic of Science over Eliezer’s essays on Bayesian probability theory. However, in my experience, the Sequences offer a special suite of mental skills and stances I haven’t found elsewhere.
Similarly to TurnTrout’s point about the sequences, learning logic/computation/ML is certainly relevant to and useful for AI safety, but there are things in Superintelligence which no computer science textbook will tell you. It’s certainly valuable to pick the most useful resources within whatever field you’re trying to study, but picking your field based on which one has the best textbooks seems misguided.
Also, textbooks typically require a great deal more effort than popular science books, so if OP is struggling with motivation for the latter, textbooks are likely to make things worse.
These works are written for a popular audience, they only teach you to talk the talk. I think it’s better to read textbooks and solve exercises, that way you learn to walk the walk. For some topics, like AI risk, there aren’t any textbooks with exercises; but you’ll still do good by learning adjacent topics like logic/computation/ML, for which there are good textbooks.
A good strategy is to read a chapter, then solve all exercises not marked “very hard” before moving to the next. Otherwise—no reading ahead. If some exercise is blocking you, you can peek at the answer, but only after spending 5 uninterrupted minutes trying to solve the exercise.
If you’re trying to learn probability theory, I think you’d indeed be better off with Jaynes’ Probability Theory: The Logic of Science over Eliezer’s essays on Bayesian probability theory. However, in my experience, the Sequences offer a special suite of mental skills and stances I haven’t found elsewhere.
Similarly to TurnTrout’s point about the sequences, learning logic/computation/ML is certainly relevant to and useful for AI safety, but there are things in Superintelligence which no computer science textbook will tell you. It’s certainly valuable to pick the most useful resources within whatever field you’re trying to study, but picking your field based on which one has the best textbooks seems misguided.
Also, textbooks typically require a great deal more effort than popular science books, so if OP is struggling with motivation for the latter, textbooks are likely to make things worse.