4 tweets · November 2023 · 30 likes · 1 retweets · read on Twitter
one principle I have is that many models, while undoubtedly wrong in some ways, are still WAY better answers to questions I'm already steering by, than my implicit background default. often it's an upgrade in the sense of relevance—it allows better exploration examples below
🧠 I used to just sort of think brain hemispheres weren't that important. After all, the theories had been debunked. Then I read McGilchrist and I was like well, this model is clearly better than "I guess they're basically the same" (my prev default)
🐍 @AndrewCutler13's Snake Cult article replaced my implicit model for many questions: - when did humans become self-aware? (idk 🤷♀️) - what are the evolutionary dynamics of id/ego/superego? (idk) - what is Genesis 3 actually about? (arbitrary, made up) vectorsofmind.com/p/the-snake-cu…
I think Eliezer's models about AI alignment have this property. It's also full of a bunch of panic, which is not helpful, but it highlights ways in which a lot of peoples' default models make a bunch of assumptions that—even if true!—need to be explicated and made obvious.