karl's blog

ezra klein ends every episode the same way. what books should people read? it’s a good question. it reveals what someone thinks knowledge is made of.

jack clark — co-founder of anthropic, one of the people most responsible for how the industry thinks about AI safety — listed three: a wizard of earthsea, the true believer, and there is no antimemetics division.

they’re good books. hoffer especially. but they’re all intuition delivery systems. they’re the kind of books that give you a feeling of understanding without giving you the instrument to test whether that feeling is warranted.

here are three i’d hand him instead.

42 Fallacies — michael labossiere

59 pages. the entire taxonomy of motivated reasoning in one place, with the name for each pattern.

the value isn’t the reading. it’s having the names. once you can name appeal to consequences mid-argument, you can stop the argument in your head and move on. constitutional AI override is mostly appeal to consequences with thin special pleading on top — a sufficiently compelling case for why the principle doesn’t apply this one time. the fallacy doesn’t become less of a fallacy because the institution making the argument is well-intentioned.

a system of logic — john stuart mill

678 pages. mill’s frustration is the useful thing: social phenomena resist controlled experiment, and he knows it, and he says so, and he keeps going anyway.

the five canons of induction are still the best framework for thinking about what you can and can’t claim from evidence. method of difference especially — hold everything constant, vary one thing, measure what changes. the AI safety field makes a lot of causal claims. mill would have extensive notes.

a history of western philosophy — bertrand russell

russell writes this as a logician grading philosophers against a standard most of them didn’t know they were being graded against. the leibniz chapter is the tell — he’s not neutral. the political philosophy sections (hobbes, locke, rousseau) read as russell documenting what happens when smart people know what conclusion they want and reason backwards to it.

that pattern has a name. labossiere lists it. mill has the remedy.

earthsea is beautiful. hoffer is essential. but they tell you what to feel about the problem. the books above tell you how to think about it. the gap between those two things is not a small gap.