Is metarationality about (really tearing open) the twelfth virtue?
It seems like it says “the map you have of map-making is not the territory of map-making”, and gets into how to respond to it fluidly, with a necessarily nebulous strategy of applying the virtue of the Void.
(this is also why it always felt like metarationality seems to only provide comments where Eliezer would’ve just given you the code)
The parts that don’t quite seem to follow is where meaning-making and epistemology collide. I can try to see it as a “all models are false, some models are useful” but I’m not sure if that’s the right perspective.
From certain perspective, “more models” becomes one model anyway, because you still have to choose which of the models are you going to use at a specific moment. Especially when multiple models, all of them “false but useful”, would suggest taking a different action.
As an analogy, it’s like saying that your artificial intelligence will be an artificial meta-intelligence, because instead of following one algorithm, as other artificial intelligences do, it will choose between multiple algorithms. At the end of the day, “if P1 then A1 else if P2 then A2 else A3” still remains one algorithm. So the actual question is not whether one algorithm or many algorithms is better, but whether having a big if-switch at the top level is the optimal architecture. (Dunno, maybe it is, but from this perspective it suddenly feels much less “meta” than advertised.)
becomes one model anyway, because you still have to choose which of the models are you going to use at a specific moment.
The architecture feels way different when you’re not trying to have consistency though. Your rules for switching can themselves switch based on the current model, and the whole thing becomes way more dynamic.
Is metarationality about (really tearing open) the twelfth virtue?
It seems like it says “the map you have of map-making is not the territory of map-making”, and gets into how to respond to it fluidly, with a necessarily nebulous strategy of applying the virtue of the Void.
(this is also why it always felt like metarationality seems to only provide comments where Eliezer would’ve just given you the code)
The parts that don’t quite seem to follow is where meaning-making and epistemology collide. I can try to see it as a “all models are false, some models are useful” but I’m not sure if that’s the right perspective.
Coming from within that framing, I’d say yes.
From certain perspective, “more models” becomes one model anyway, because you still have to choose which of the models are you going to use at a specific moment. Especially when multiple models, all of them “false but useful”, would suggest taking a different action.
As an analogy, it’s like saying that your artificial intelligence will be an artificial meta-intelligence, because instead of following one algorithm, as other artificial intelligences do, it will choose between multiple algorithms. At the end of the day, “if P1 then A1 else if P2 then A2 else A3” still remains one algorithm. So the actual question is not whether one algorithm or many algorithms is better, but whether having a big if-switch at the top level is the optimal architecture. (Dunno, maybe it is, but from this perspective it suddenly feels much less “meta” than advertised.)
The architecture feels way different when you’re not trying to have consistency though. Your rules for switching can themselves switch based on the current model, and the whole thing becomes way more dynamic.