I think it’s best to view weirdness points as a fake framework.
I don’t think there is, at any level of abstraction, an accurate gears level model that includes weirdness points as a gear. But, if you’re just trying to make quick and dirty heuristics about what you can get away with, it’s an excellent heuristic.
When you’re looking at the gears of this phenomena, I think you start looking at signaling and countersignaling, which will give you more accurate answers than trying to count weirdness points.
A model that makes accurate predictions at a given level of abstraction. and can handle may cases at that level E.g. if the level of abstraction is “human behavior” (rather than say, quarks) it should give accurate predictions about the human behavior abstraction.
What you are talking about is a function of given-level-of-description, not an absolute. So there is a level of abstraction where “weirdness points” works.
Probably, but not a very useful one. It’s better to just use natural levels of abstraction like “human behavior” and recognize that this is not a gears level model for that level, but a heuristic. I can’t really think of a natural abstraction where weirdness points is usefully gearsy, rather than a heursitic.
Gears level is defined by prediction power at a given level of abstraction, heuristic is defined by something like… “speed/prediction power” at a typical level of abstraction, or something. Whether you want gears or heuristics really depends on how much time you have and how much time you’re going to spending with the model (typically heuristics).
I think it’s best to view weirdness points as a fake framework.
I don’t think there is, at any level of abstraction, an accurate gears level model that includes weirdness points as a gear. But, if you’re just trying to make quick and dirty heuristics about what you can get away with, it’s an excellent heuristic.
When you’re looking at the gears of this phenomena, I think you start looking at signaling and countersignaling, which will give you more accurate answers than trying to count weirdness points.
Given that you can’t have a quark level model , what counts as a gear level model?
A model that makes accurate predictions at a given level of abstraction. and can handle may cases at that level E.g. if the level of abstraction is “human behavior” (rather than say, quarks) it should give accurate predictions about the human behavior abstraction.
What you are talking about is a function of given-level-of-description, not an absolute. So there is a level of abstraction where “weirdness points” works.
Probably, but not a very useful one. It’s better to just use natural levels of abstraction like “human behavior” and recognize that this is not a gears level model for that level, but a heuristic. I can’t really think of a natural abstraction where weirdness points is usefully gearsy, rather than a heursitic.
“gears level” is defined in terms of usefulness, and so is “heuristic”
Gears level is defined by prediction power at a given level of abstraction, heuristic is defined by something like… “speed/prediction power” at a typical level of abstraction, or something. Whether you want gears or heuristics really depends on how much time you have and how much time you’re going to spending with the model (typically heuristics).
Are they really different? Why would you want to use a high abstraction level if not to save Compute?
Yes.
You do want to use it to save compute.
So why are they different?