I endorse and operate by Crocker’s rules.
I have not signed any agreements whose existence I cannot mention.
Mateusz Bagiński
Karma: 3,822
EM
?
Likewise if you ask me what day is 4 days from now, the thing I automatically do is seek forwards until I’ve moved 4 spaces.
I also do that with “small shifts”: I imagine taking 1-3 steps forward/backward or something. For ≥4, I calculate (June=6, so 5 months from now is 11=November).
What’s time within a day like for you?
It’s a number line, but a jagged one. 3:30 to 8:30-ish, it’s quite steep, like 30° or maybe even 50°. Then it’s more flat (but still monotonically rising) until like 14:30, when it starts rising a bit more rapidly, slows down again at 20:00, and then it crosses to the next day. Full hours are marked with a mental flag, and particular orientational times, like when I typically eat meals or wake up or go to sleep, are marked with bigger mental flags. Time within an hour or a minute is similar but less detailed, as I don’t think about it / zoom in on this often; plus there’s the feeling of a container slowly filling up with time, like 14:58 is veeery close to the tipping point.
(The way I described the day feels kind of off because in my mind’s eye, it’s more cloudy/blurry/fuzzed, and it’s plausible that there’s some significant variance in how I feel it.)
I perceive the historical timeline similarly. Replace the day with a decade, a century, a millennium. It’s slowly, but unevenly, rising upwards, with bigger steps at the boundaries of millennia, centuries, and decades.
I don’t have much mental imagery for months. Too irregular. If I want to know the number of days to a certain date or what day of the week it would be, I count.
I think of that just via the numbers, and was surprised the first time I heard someone tell me that obviously analog clocks were easier because you add the angles on a circle in your head. (This is probably a generational thing: I grew up with digital clocks, and had to be taught in school how to read an analog clock like an extra skill, and would only have a reason to need that skill if I was in class and wanted to surreptitiously check how long it’d be until it was over).
Yeah, I can see that working for some people. I guess I grew up with a mix of analog and digital (and there’s still a lot of analog clocks around me (is it a Europe thing?)), but never adopted the former as a privileged object in my imagery.
Also, this reminded me of a fun thing: https://faculty.ucr.edu/~eschwitz/SchwitzPapers/DreamB&W.pdf
In the 1950s, dream researchers commonly thought that dreams were predominantly a black and white phenomenon, although both earlier and later treatments of dreaming assume or assert that dreams have color. The first half of the twentieth century saw the rise of black and white film media, and it is likely that the emergence of the view that dreams are black and white was connected to this change in film technology. If our opinions about basic features of our dreams can change with changes in technology, it seems to follow that our knowledge of the experience of dreaming is much less secure than we might at first have thought it to be.
I’ve recently finished running the first AFFINE Superint Alignment Seminar, which went quite well and led me to discover some promising people for whom that amount of money would make a huge difference at this point.
I’ll contact you over DMs.
Would be nice. Any idea how to make it more likely?
One thing to do would be to develop an airtight (partly technical?) argument for ~doom on the current trajectory or whatever, and then publish it in a format that is understandable (if partly through deference) by the relevant crowd at the companies, as well as the relevant crowd outside of the companies (investors?), so that clearly the only socially acceptable move is to shut down or something. (Some lobbing at the margin will plausibly be required.)
I’m also interested in historical examples of companies shutting down for vaguely analogous reasons. Has any tobacco company shut down after it became common knowledge that smoking is bad?
What’s your estimate of technical alignment research FTEs at MIRI at the time of your departure?
Partitions are categorically dual to both subsets and factorizations, kinda
[Prerequisite warning: basic category theory.]
In his work on Finite Factored Sets, Scott Garrabrant defined the notion of factorization. In simple terms, a factorization of a finite set
Scott wrote:
Notice the duality between the definitions of partition and factorization. We replace subsets with partitions, nonempty with nontrivial, and disjoint union with Cartesian product, and we reverse the direction of the function. We can think of a factorization of S as a way to view S as a product, in the same way that a partition was a way to view S as a disjoint union.
Intuitively, this makes sense, and I tried to put a bit more categorical flesh on this in my comment under that post. In the category Set (or rather FinSet, as we’re only dealing with finite sets), the categorical product (
Factorization:
Partition:
(Point of order: perhaps, the
But there’s an alternative way to view a partition, which makes things more complicated. We can think of a partition of
So, depending on how we represent a partition of
I was trying to put subsets, partitions, and factorizations into a unified picture (which wouldn’t require us to switch between the ways of representing the partition), but without much success. It stands out to me that there seems to be a hierarchy of structures partially ordered in some natural ways:
Subsets — ordered by inclusion (
Partitions — formed from non-trivial (non-empty) subsets — ordered by from the finest to the coarsest (
Factorizations — formed from non-trivial (non-singleton) partitions — ordered by history (?)
Is there something interesting “above” factorizations that can be built from non-trivial factorizations, where a factorization is trivial if it contains only one partition (?) (though that should be eliminated by the “formed from non-trivial partitions” requirement?)?
We could add elements of
Another thing maybe of interest: We can check whether two subsets can be included in the same partition by checking if they’re disjoint:
Any suggestions/observations on whether there’s some “there” there much appreciated!
- ^
Possibly this is a mistake, though. Possibly by demanding that the
The last obstacle I was working with is, what is equality for sets? Classically, two sets are equal if they contain the same elements. I basically need to keep that because I don’t think one can do mathematics without it, but when you have non-well-founded sets, it becomes a lot more involved.
A first-pass category-theoretical answer is that sets are equal iff they’re isomorphic.
You get a more interesting answer if you identify them both as subsets of a larger set, connected via inclusion morphisms. Then you obtain their intersection by pull-back-ing one inclusion along the other. If the projections of the intersection are isomorphisms, then they contain the same elements / their inclusion morphisms are the same.
There’s an interesting line in the old MIRI paper on Definability of Truth: “Tarski’s result on the undefinability of truth is in some sense an artifact of the infinite precision demanded by reasoning about complete certainty.”
Not to my knowledge (though I haven’t searched), but it’s funny that Adrià and Abram found a solution by kind of imposing even more precision via adding hyperreal numbers: https://www.lesswrong.com/posts/j3Cv94eEZTDFiLREW/infinitesimally-false
An alternative idea (inspired by some of davidad’s thinking) is that intelligent minds might create logical timelike and spacelike separation as a consequence of their decisions. Demski’s conception of logical time (which was anticipated by Lacan!) is mostly motivated by the idea of agents recursively reasoning about each other. It seems like such agents could create a kind of directionality by simply refusing to reason in certain ways, as part of a correlated game-theoretic strategy. (As an aside, the idea of refusing to do certain kinds of reasoning about other agents seems like an important building block for defining concepts like respect, trust, and faith.) This could lead to a logical analogue of the lightspeed limit which is less of a hard constraint, and more like a highway speed limit: breakable, but in a way which might incur larger consequences (specifically, exclusion from the coalition that coordinates itself by imposing spacelike and timelike structure in the logic of rational minds).
[Epistemic status: babble.]
I’ve been reading Schelling’s Strategy of Conflict recently, and I learned that the concept of Schelling points is much richer than I expected and extends beyond tacit (~acausal-ish) bargaining/cooperation to contexts with communications channels, in which Schelling points (among other things?) give structure (e.g., boundaries) to the space of possible bargains, e.g., how high should I try ramping up the price before stopping or what values should I consider as options for my bidding? (Maybe there’s some connection to your past shortform on boundaries?)
If we generalize the space of possible bargains to the space of various logical facts (or their representations in the minds of agents), we arrive at the possibility that a process like this one might give space-time-like structure to the space of logical facts.
More broadly, this might kinda favor a constructivist/computationalist approach to math/logic, where the right way to think about it (at least for the purpose of finding space-time-like structure in logical spaces) puts a lot of weight on fixed points / equilibria that recursively reasoning agents can be broadly expected to converge on.
If so, another analogy that comes to mind is that between equilibrium and non-equlibrium reasoning, e.g., in economics, evolution, etc., where the idealized equilibrium is a boring state where not much happens, with stuff actually happening in non-equlibrium settings (see, e.g., complexity economics, maybe ergodicity economics?). You get to the equilibrium in the limit assuming the relevant sort of structure in which the process proceeds (analogous to the loss landscape in SGD or the evolutionary landscape in biology) stays fixed. But it may take a lot of time to get to the equilibrium and the relevant structure doesn’t, in general, stay fixed. The connection is that assuming logical omniscience or unbounded computational resources is that sort of in-the-limit reasoning. This sort of reasoning sometimes flattens out some relevant sort of structure, and we can re-derive this structure by expanding the domain of discourse.
My little brother was accused by his preschool friends of making things up when he was trying to tell them that in a billion years the sun would turn into a red giant and consume the earth.
Why would someone work (/ pay for working) on options for possible future coordination around a pause if they disapprove of the pause? What would they expect to gain?
Also, even developing options without endorsement (/ with dis-endorsement) indicates that one considers pause potentially viable in the future, which is a massive change from the situation, e.g., one year ago, when Anthropic and OpenAI clearly considered it not worth their thought.
For completeness, a cute factoid is that if we view monoids and groups as categories (that is, [categories with one object] and [categories with one object and every arrow reversible], respectively), then the move from one object to multiple objects creates groupoids from groups and categories from monoids. Hence, categories are monoidoids.
The apparent existence of qualiagnosia has made me update positively on eliminativism/illusionism. But, a possible rejoinder could be that there are genuine qualia-havers, and genuine qualiagnosics (the two other options are p-zombies, and people who have qualia but say they don’t have them).
It might also be that qualiagnosics have qualia but don’t know it, similarly to how aphantasiacs actually have various forms of imagination, largely similar to non-aphantasiacs’, except mostly running in the background.[1]
If it ends up that there is a biological foundation of qualia-reporting/qualiagnosia, there’s some interesting ethics. Not that one shouldn’t treat qualiagnosics as moral patients (and most moral theories agree, though how exactly hedonic utilitarianism would justify that is unclear). Tbc I think we should treat qualiagnosics as moral patients.
From my understanding Tomasik is both an eliminativist and a hardcore negative utilitarian, so I’d guess he has some takes on this, although they might mostly reduce to something like: suffering defined as a computational pattern similar to whatever we call suffering in humans.
- ^
I consider it coherent to have qualia and be unaware of the fact, insofar as we’re taking qualia to be a good concept.
- ^
It takes a lot of time
IME it’s like 10 seconds for a room of 50 people?
isn’t explicit about what it’s doing
I think I first experienced it ~4 years ago, and it quickly became obvious to me what it’s trying to do.
Another example is https://gwern.net/backstop
A foundation for many nested search/optimization/adaptation processes is that the outermost loop has a faithful access to a right sort of “ground truth signal”. This tends to be inefficient/costly/slow but nevertheless necessary because without it the more efficient/greedy but less sensitive(/wise?) inner-er processes tend to go awry and the whole optimization edifice collapses or maybe grows cancer or something.
Socionics is kinda MBTI-adjacent, but has more interesting, fleshed-out structure, with systematic predictions, e.g., predicting synchronies or conflicts between personality types that are related in some specific way.
make the one true categorization of Types of Guy
Do you want a compact description, e.g., some small-ish number of naturally discretizable factors, with perhaps many combinations being very sparsely populated, and that would at the same time comprehensively 80⁄20 a person’s personality description? I would not expect a comprehensive theory of human personality to be so neat.
The way I’d phrase it is that intelligent people often act dumb, unless they learn to act smart.
Yeah, niplav also brought my attention to this.
Slightly upweights the hypothesis that I “just” got explicitized insight into how temporal cycles are “deeply represented” (in some relevant sense) in my brain.
I don’t understand why you believe this and I don’t see how the rest of your post supports this position or even how it might illustrate that this position makes sense.