i am reginald reagan aka RAGIN' RAYGUNS

aka Dr. Molecules aka Mr. Statistical-Probabilities but you can call me Reggie (though it's a pseudonym)

[raginrayguns]

Payorian cooperation is easy with Kripke frames — LessWrong

The context is MIRI's twist on Axelrod's Prisoner's Dilemma tournament. Axelrod's competitors were programs, facing each other in an iterate

Finally wrote this up after a conversation with @youzicha, who also helped throughout the whole writing process this past week.

In 2023 I began learning modal logic from a book recommended by @eka-mark, Graham Priest's "An Introduction to Non-Classical Logic", which also happens to be a very good introduction to classical logic. I was hoping to formalize the thoughts about decision theory I had been posting on Tumblr. I really liked the tableau method used in the book, which I now understand to be a formal system that sticks closely to Kripke semantics. In contrast to the natural deduction systems they use for modal logic on LessWrong, I was learning an algorithm, a decision procedure.

I thought MIRI's prisoner's dilemma tournament would be a good tutorial level to try out the technique. Instead, it felt more like a final boss, which was humbling.

But I recently stumbled across an Andrew Critch post that suggests a different modal agent for the tournament. When I tried the tableau technique I had practiced, it all just worked, and easily.

This post is the tutorial level I had hoped for, applying Kripke semantics to "modal combat". It's technical, but I think it's doable if you know what provability logic is and you've read the MIRI prisoner's dilemma paper.

The exercise has restored my confidence that an amateur can productively think about these topics. So I'm ready to get cocky again, and I hope I'll have more to say soon.

[raginrayguns]

I'm not saying that the Chinese government should support a monarchist coup in the UK if King Charles promises not to nationalize British Steel, but it would be only fair

[eka-mark]

It's awfully considerate of nonempty finite sets to have an equal number of odd-sized and even-sized subsets.

(Hint: apply the binomial theorem to (-1+1)^n.)

[raginrayguns]

or, by induction on pascal's triangle: every number contributes equally to two adjacent numbers below it, one odd-indexed and one even-indexed

[raginrayguns]

or, after remembering why pascal's triangle counts subsets of each size...

when you add a new element to a set, you keep each of the old subsets, but you also get a new subset: the old subset, with the new element added.

so every old subset contributes two sets: itself, and itself with the new element.

One of these is of odd size and the other of even size.

And that's it, we actually don't need induction

it's still important that we created the set by adding a new element to a smaller set, since that's why this reasoning doesn't apply to the empty set (which has an even-sized subset of size zero, and no odd-sized subsets)

[argumate]

I must have seen dozens of fictional astronauts come to grief in EVAs gone awry and float away into eternity when this has literally never happened in reality, although I did read this terrifying story of an Italian astronaut on the ISS:

In 2013, Luca Parmitano experienced one of the most dangerous EVA emergencies ever. During a spacewalk, water began leaking into his helmet from the suit's cooling system. The water eventually covered his eyes, ears, and nose, making it difficult to see or hear. He had to navigate back to the airlock partially blind. Had the leak continued longer, drowning inside the helmet was a real possibility.

I thought space was the one place you wouldn't drown, damn

[raginrayguns]

Was wondering how it could cover his eyes and nose but not his mouth, his mouth is lower down... oh, right, they're in orbit

[eka-mark]

It's awfully considerate of nonempty finite sets to have an equal number of odd-sized and even-sized subsets.

(Hint: apply the binomial theorem to (-1+1)^n.)

[raginrayguns]

or, by induction on pascal's triangle: every number contributes equally to two adjacent numbers below it, one odd-indexed and one even-indexed

[defendglobe]

project hail mary but grace's translation program uses that awful fucking default tiktok TTS voice for rocky instead.

[argumate]

one thing LLMs and people have in common is if you ask them "why did you do that" even their sincere answer will often be complete garbage confabulated after the fact.

[raginrayguns]

Piranesi seems to me like the perfect execution of the present sff "formula": a protagonist that inspires sympathy, a world that evokes curiosity, the perfectly timed drip of information about it, the steady ratcheting up of suspense to a confrontation at the climax.

i think it's also the only fantasy novel written by an author that debuted in this century that i've really fully liked, without some deep frustration with it

also just one of the best ever british fantasy novels in the style of tolkien and lewis, like... idk, there's something distinctive about it, like for better or worse the author seems to some extent to believe in this stuff.

And it lost the Hugos to a murderbot novel?

what is murderbot? Either I'm missing out or i'm mad at the sff fandom

or maybe it's just weird that these share a category and it has more to do with the relative size of audience for very different kinds of things... ?

[raginrayguns]

I read the first chapter of network effect, it seemed like a very standard capefic opening. I don't really like capefic. Not that it's capefic as in they wear spandex and capes, but i guess it's the same kind of appeal? So... not my thing, have fun guys

[raginrayguns]

@manyblinkinglights said:

why wouldn’t you start with the first one… the rest of them are like. “Well I guess you liked Murderbot” and we’re all like “we did!!!” but the first one was the, like, fresh one.

cause network effect won a hugo award. It beat piranesi in fact

[raginrayguns]

i feel like "lonely girl becomes friends with supernatural spirits" is a common fantasy setup and what I liked about Among Others was that it really follows through on the spirits not really being a substitute for human friendship. Like in Pale or Gunnerkrigg Court I almost feel it undermines the premise somewhat that the spirits are fully human, rather than less than human, or at least missing much of what we require from humans

[raginrayguns]

Piranesi seems to me like the perfect execution of the present sff "formula": a protagonist that inspires sympathy, a world that evokes curiosity, the perfectly timed drip of information about it, the steady ratcheting up of suspense to a confrontation at the climax.

i think it's also the only fantasy novel written by an author that debuted in this century that i've really fully liked, without some deep frustration with it

also just one of the best ever british fantasy novels in the style of tolkien and lewis, like... idk, there's something distinctive about it, like for better or worse the author seems to some extent to believe in this stuff.

And it lost the Hugos to a murderbot novel?

what is murderbot? Either I'm missing out or i'm mad at the sff fandom

or maybe it's just weird that these share a category and it has more to do with the relative size of audience for very different kinds of things... ?

[raginrayguns]

I read the first chapter of network effect, it seemed like a very standard capefic opening. I don't really like capefic. Not that it's capefic as in they wear spandex and capes, but i guess it's the same kind of appeal? So... not my thing, have fun guys

[icecream-eaterrr]

I just heard this woman say “you procrastinate because you are afraid of rejection. It’s a defense mechanism, you are trying to protect yourself without even trying.” and I think I just realized what was wrong with me.

[raginrayguns]

In Savage's decision theory, acts are functions that map states of the world to consequences. Since he also defines "the world" as "the object of interest", for example an egg, I'd rather say "states of the system".

Consider a physical example. A chamber which may or may not have an electric field; maybe we can switch it on or off. So our "states" are on or off. Our acts: we can shoot a positive particle through it, or a negative particle. So one act maps "off" to "goes straight", and "on" to "curves up". The other act maps "off" to "goes straight", and "on" to "curves down".

These "acts" are just the time evolution of the physical system. The "state" of course is not really a full description sufficient for the time evolution. In this case, though I think this is inessential, the "state" specifies the laws, and the initial conditions depend on the act. One act is time evolution from the initial condition of a positive particle going through the chamber, and the other act is time evolution from a different initial condition where the particle is negative.

The reason this kind of physical interpretation of Savage's decision theory appeals to me is because it seems to describe my experience with "virtual screening", where I did quantum chemistry simulations to decide which organic molecule to synthesize. Like, let's say we're considering two molecules, anthracene and alizarin, and we want one that absorbs blue light, making it a red dye. We're at a dye company in the 19th century but with computers or something, I don't know. Although this doesn't really correspond to how I actually do the simulations, we could think of this as before, with acts corresponding to time evolution under different initial conditions, but with the same Hamiltonian. The Hamiltonian will be an oscillating electric field with the frequency of blue light, and the consequences are that the energy of the system goes up (absorption) or not. We don't have multiple states here, because there's no uncertainty in this decision problem.

I also think this kind of physical interpretation is philosophically valuable because it tells you something about the scope of the decision theory, where it will and will not work. It only really makes sense with a system that's isolated enough that (given a state) it responds predictably to our action. But it's not a theory for a completely isolated system, since the initial conditions get set outside the scope of the explicit model. I think this is not obvious, or at least wasn't to Savage, and makes it less surprising when you can't "internalize" the agent without running into issues like Newcomb's problem.

[alexanderwales]

I was reading through the U.S. Copyright Office's "What Does Copyright Protect?" page, mostly to make sure that I wasn't being wrong on the internet, but I absolutely love that this is part of their FAQ:

How do I protect my sighting of Elvis? Copyright law does not protect sightings. [,,,]

There are only 12 questions on the FAQ, and this is one of them.

[wildgifthorses]

Can I get a star named after me and claim copyright to it?

No. There is a lot of misunderstanding about this. [,,,]

[raginrayguns]

One fun thing I remember from Nozick's original paper on Newcomb's problem is undermining the dominance argument by "swapping" the elements of the 2x2 decision table.

Like, suppose we have states s₁ and s₂, and actions a₁ and a₂:

Suppose consequence B is preferred to A, and Y is preferred to X, so a₂ dominates.

We could instead write the table in terms of these states:

So to Nozick, this "vertical relationship" in the original table, B being under A and Y under X, is arbitrary.

But that's because he thinks of a₁ and a₂ as propositions, as descriptions of the system. To get into the classical decision theory mindset, you have to have a pretty limited system in mind, which you always do in the context of scientific modeling.

[dragon-in-a-fez]

one of the funniest conversations I ever had with my ex was when they were still getting used to Celsius and asked me "what's 20 degrees?" and instead of converting it, I said "it's the highest your dad will ever let you set the thermostat and when you say you're cold he tells you to put on another sweater, we're not made of money" and they went "oh, 68"

the fact that this reference was that fucking precise was something they went on to tell people about for years.

[raginrayguns]

one thing i see a lot that doesn't land with me the way it used to, is when i see someone trying to get people whipped up about speculative research direction or even about a completely hypothetical field of science

the more i do research the more worthless such things seem

on the other hand, even a little bit of substance can be very exciting. Like, one model or relationship. I guess this is Kuhnian; "paradigm" means example, his idea was that ordinary scientific research in a field is a sort of imitation of one foundational success

[raginrayguns]

looked a bit at dan dennett breaking the spell, looks like another case of this?

[raginrayguns]

@urpriest Replied to your post “Sam Harris writing If the Palestinians simply...”:

For the Dan Dennet point, I think he was a lot more interested in anthropology-flavored comparative religions than neuroscience-flavored stuff. Certainly when he debated theists it was the former that he emphasized, stuff like "look how this modern-day religion started, why do you think yours started differently?" rather than Sam Harris's stuff.

Yeah, I looked in Breaking the Spell, and he says:

Until we develop better general theories of cognitive architecture for the representation of content in the brain, using neuro-imaging to study religious beliefs is almost as hapless as using a voltmeter to study a chess-playing computer.