Check out my website: thermontology.com
interstice
Karma: 2,998
“Messiness” seems to pop up only when you try to interpret domain A through domain B, but you’re confused about how to do it (e.g. you don’t know statistics) or lack capabilities to do it (e.g. you don’t have precise enough instruments).
Hmmm, so I think of messiness as being like...essentially, since the world is complex, the best models of most phenomena have many parameters.
But this means that, for long-running disagreements between world-models, most pieces of evidence won’t be able to resolve them. Because each world-model will have a lot of free parameters to interpret a new piece of evidence in any given way. This can be especially bad for controversial topics since people will engage in motivated reasoning, which there’s plenty of room for due to the abundance of free parameters. It’s also computationally expensive.
Thus to distinguish between world models, it’s often actually easier if the data comes from a small, artificially restricted domain so that there’s less wiggle room.
As an example, consider the “pots VS peoples” theories of the indo-europeans. That was definitively resolved by DNA data. Though people had plenty of archaeological data, it was not as disambiguating.
your definition doesn’t define what is an interesting question/problem
I think of interesting problems as being those whose answers (you think will) most help you disambiguate your global world model.
Gettier cases....yes there’s some similarity in examining an edge case. Lots of philosophical thought experiments are like that(but to be honest I never really understood the intuition that the weird JTB cases were not “knowledge” or why this distinction was important)
I actually think of “questions” and “problems” as being fairly neat concepts. “Question” just being some thing you can query about the system, like “what is the position of Mars tonight?”, “what is the regulator of this elliptic curve?”, whereas “problems” would be searches over easily verifiable structures, like “does there exist a proof of this conjecture?”, “does there exist a way of synthesizing such and such a chemical element?”
Regarding ontology...I think of flywheels as being small, toy domains which give unusually clear, unambiguous answers, which can aid in the construction of theories applicable to messier domains. “High resolution” parts of reality, if you will. Take the movement of the planets: the laws of motion can be seen more easily when separated from the vagaries of Earthly existence such as air drag. Or particle accelerators are literally higher resolution than most things(they probe higher energy levels ~ smaller scales). This is connected to my overall schema of hot VS cold things which I’ll get into later.
I’d like to frame philosophical undertaking in a way that highlights practical applications and consequences
What do you think of my discussion of usefulness in 1.6? I agree my discussion is mindy but I feel like all philosophy is mindy(even the purportedly anti-mindy variety)
I think of a flywheel as being something that is easy to iterate on and get unambiguous results. So yes, basically “well defined measurements” + “smaller state-space”.
Combining two flywheels into one...hmm, maybe. I would be a bit more general and say that deep discoveries compress a lot of your worldview, which could include uniting two small well-defined domains(or showing a particular small domain well predicts a bigger domain)
Funnily enough I just wrote a post about empirical flywheels in math and philosophy which I think are ~the same thing you call “scrying”.
I think you maybe could model it in logical induction? Basically when you are “scrying” you’re working in some sort of small verifiable domain which can give you lots of unambiguous feedback quickly, in the hopes that the information you get will be applicable to some larger setting you care more about. Doing it well would be like, developing good intuition/heuristics about what sort of small verifiable problems tend to be useful to work on.
Sure. By “not explicitly pre-trained” I just mean to say that there’s nothing ‘special’ about the characters from the training algorithm’s point of view, so in this respect they’re not so different from a hypothetical general predictive algorithm in humans(although actually I guess the human brain attaches special salience to other people, but regardless...)
It has a general objective of next-token prediction for which modeling characters is a useful strategy. IMO it’s plausible that the human brain is “trained” on prediction to a large extent, for which modeling characters is also a useful strategy.
Well technically speaking AIs aren’t explicitly pre-trained to predict what characters would do either, characters are an emergent feature extracted from next token prediction.
There are definitely some differences(AIs get way more character pre-training data, there’s an explicit separation of pre- and post-training, humans have lifetime memory) but overall I think the “agentic part is a subset of a predictive model” thing is pretty plausible in both cases.
These don’t feel that contradictory to me. You could think of the ‘persona’ as being the main agentic actor in the system. Possibly to be replaced by something more sophisticated when AGI is invented, but maybe not? GPT5.5 and Fable show that persona intelligence can get very high. I’d say it’s even plausible that humans are personas, in the sense that the agentic part of a human is a subset of a general predictive world model. This is one way of interpreting some meditative experiences of “dissolving the self”.
@Wei Dai you might find interesting?
Although if you have very short timelines(or even a moderate probability thereof) it might not make sense at this point to invest effort in legible things that aren’t directly on your subjectively most promising path to impact.
Can belief webs reach the best equilibrium in principle? By default it seems like they might just get stuck in local equilibria: unlike FixDT they don’t have a mechanism to “jump” into the best equilibrium
Maybe you want your belief web to be high-dimensional compared to the intrinsic dimensionality of the decision problems you’re trying to solve(whatever that might mean), so there’s always some room to wiggle towards a better equilibrium.
Elegance is also related to description length, so really any reasonable prior will have to incorporate it so some degree.
I believe you can still make good money doing either competently.
Are you familiar with Robin Hanson’s work on hard steps in the development of life and grabby aliens? Summarized e.g. at the beginning here.
I’m making a blog: thermontology.com The theme is going to be the relationship(or lack thereof) between the laws of physics and high-level structure in our world such as intelligent life. Sort of a step towards what Vanessa kosoy calls “metacosmology”. I’m also going to stake out a possible stance towards “metaphilosophy”. People here might be interested in the topics so I’m probably going to cross post a bunch!
I’ve felt similarly for years, many posts somewhat worth skimming but only a few worth reading in depth. You may just have absorbed most of the novel bits in the local memeplex.
Yes. But the OP is about contrasting people like Paul with Eliezer. Paul (I think) does indeed predict a dramatic singularity, but also that the ramp up to said singularity will be smoother and more widely distributed across society than Eliezer predicts.
I think the heuristic “nothing ever happens” is better interpreted to mean “nothing ever happens relative to baseline trends” than “literally nothing ever happens”. The incrementalist worldview seems like a better fit for this heuristic than Eliezer’s, which after all ultimately predicts something very dramatic happening.
OT, but what do you think of my take that philosophy is about mesa-optimizers extracting patterns (relevant to their functioning) from their underlying statistical model into a legible format?