You may remember me spotlighting the Mawasi Al-Qarara Mutual Aid Project (MAQMAP) in the past. MAQMAP is now Relief for Rafah (R4R), as the organisers and inhabitants of the camp return to their homes across Gaza.

I am personally organising bank transfers for this organisation and have been in personal contact with the organisers for some time. I can assure you that your donations are doing important work in Rafah.

i caved and made one of those

Okay I did one of the things

my favourite metaphors tend to be physical/computer science ones: the feedback loop, the dynamical system, the state space, the instability, the self-organising structure, the mapping from x-space to y-space, the nth-order approximation, the evolving population, the compression algorithm, the cellular automaton, the fractal structure, the soliton, entropy. as far back as 2018 i was conceiving of gender transition's relation to society as akin to bubbles in a fluid.

i like to spice it with some more occult stuff now and again, the egregore in particular (used in preference to similar, more atomic concepts like 'meme' or 'semiotic sign' or 'stand alone complex' mostly because i like the vibe) - but that's a flavour of occultism that suits this habit of thought, isn't it? a notional abstract entity that emerges from the dynamics of a complex system, such as multiple minds? i view magic mostly in this light: a human tool for apprehending the large scale behaviour of humans. as such my go-to examples of egregores are things like 'countries' or 'organisations' or even 'gender'.

anyway i don't think this is a bad way to look at the world, i think it often leads to interesting left field approaches to subjects, but just because i invoke all these sciencey concepts does not actually entail any rigour. I'm operating on the level of analogy and i don't want to pretend otherwise. i try to be careful to keep the technical definition in mind, but it's not like I'm writing differential equations down, or even that you could in a lot of cases... and i tend to dislike the rhetorical invocation of mathematical concepts when other people do it, i am kind of a hypocrite lol

i relate to this. a lot of my interest in phil of math is trying to understand the usage of analogies *within* math, in part bc i have an unjustifiable sense that a good understanding of them would carry over to insight into when/where mathematical analogies are useful outside of math... it's an informal phenomenon either way. i think their usage outside of math is very often just a rhetorical grab at epistemic authority, which a proper usage would have to thoroughly reject, clarifying rather than browbeating

reminds me a bit of the concept of "morality" in math

I've seen the term used in PL papers, adjacent to category theory but less strict about tone than published mathematics usually is; possibly also under the influence of this paper itself, since it came out in 2004. I always took it as somewhere in the continuum of "intuitively" to "up to isomorphism", "modulo some fiddly stuff". the sense of proofs as moral or immoral makes sense, though, fitting in with Thurston's Proof and Progress in Mathematics (which she quotes) talking about mathematicians working to increase human understanding of mathematics. a moral proof won't just demonstrate that this one statement is true, it'll give you a sense of how to think about the subject matter: how to come up with the proof yourself, how to prove other related statements.

I wouldn't say morality is quite synonymous with analogy, but analogy is a powerful way to give a "why": it can transfer an idea into a domain where you have more existing intuition, or it can let you move up a layer of abstraction by giving another example of the same abstract structure. I see why she'd equate it with category theory, since category theory at base is a framework for formalizing analogies, though I think she oversteps a bit here -- I'm not convinced all analogies are categorical or that morality all comes down to them. I'd be interested to hear a reception of the concept from someone in a different field...

on a more minor note, proof as a lowest-common-denominator of understanding is definitely an important part of its role but I'd say it's also important for verifying that your intuitive understanding is actually correct. there are plenty of historical examples of intuitive concepts having subtle flaws that were revealed by putting more rigor into the proofs, most famously in analysis (e.g. a function ought to be differentiable if it's continuous) and set theory (the various paradoxes); Lakatos's Proofs and Refutations deals with this subject through the geometric case of the Euler characteristic. category theory may be in an unusual position relative to other branches?

(Also I think the Yoneda lemma is slightly immoral.)

stupid metaphor time

transing your gender as chemical reaction. certain things catalyse it (lowering the energy barrier and making it more likely) accounting for common pretransition experiences, but don't guarantee it.

since detransition is also an option we could imagine eventually reaching a dynamic equilibrium where the number of people transitioning and detransitioning is equal!

being trans generally seems to be a lower energy state than being cis tho

anyway this fails because (among other reasons) humans are a lot more complicated than molecules (partly on account of being made of lots of molecules), and have far more equilibrium states and moves between them than chemical species do. trying to reduce it all to a single axis is Overly Reductive(TM). the fluid dynamics metaphor was better

Haha, I think you can safely blame my habit of using a different handle for each site over your memory!

It's a pleasure to share things with someone so curious and willing to experiment, I really admire how you throw yourself into learning about a topic and then write up a great explainer afterwards. (Speaking of LLM stuff, have you seen the latest Anthropic paper? Mechanistic interpretability doesn't tend to replicate very well but it feels like there's lots of potential there).

I'd love to see your take on the cast, I should have a go the old-fashioned way (drawing tablet) myself before the game wraps up. And a Jennagame of some kind is on the list to play next!

I have not, will read soon! I'm still on the prev openai paper about how their model will openly discuss cheating in its chain of thought, but reinforcement learning on the CoT just teaches the model to hide what it's doing. (It took me way too long to find that because in my head it was an anthropic papper.)

the AI-safety crowd are already all over it (zvi mowshowitz, who led me to it, called it the 'most forbidden technique'), but what I find interesting is that, before it's trained to hide, the model says stuff in its CoT like:

and then goes on to figure out that it can hack the test case instead of solving it properly. and I'm wondering like. why does it do that?

in a human, something is hard because it takes a lot of thought or difficult emotions, which takes energy, and we are living organisms who only have so much energy to go around. so we have a very natural incentive to try to find easier ways to do things. we are often 'lazy' - and that laziness has been very good for us!

so far as I know, though, when they go through reinforcement learning, AIs are not trained to do things more efficiently, just to get the right answer at the end. so why would an AI care about doing things in an easy way rather than a hard way? is this a sign that OpenAI was applying 'length of chain of thought/answer' as a training criterion to save on inference costs? or is it just picking up that this is a thing that humans sometimes do in its training set, and stochastically sometimes doing it when it's asked to solve problems? is it 'hard' in that the model is simply less likely to get to the right answer, so it looks for a more reliable method? ...tbh it's probably that actually.

I'll check out that Anthropic paper, maybe it will have some other interesting behaviours! 'what AI reveals about the structure of thought in general' is one of the most interesting aspects of this whole thing to me, so I'm excited to see what they've discovered.

I wrote ~4.5k words about the operating of LLMs, as the theory preface to a new programming series. Here's a little preview of the contents:

As with many posts, it's written for someone like 'me a few months ago': curious about the field but not yet up to speed on all the bag of tricks. Here's what I needed to find out!

But the real meat of the series will be about getting hands dirty with writing code to interact with LLM output, finding out which of these techniques actually work and what it takes to make them work, that kind of thing. Similar to previous projects with writing a rasteriser/raytracer/etc.

I would be very interesting to hear how accessible that is to someone who hasn't been mainlining ML theory for the past few months - whether it can serve its purpose as a bridge into the more technical side of things! But I hope there's at least a little new here and there even if you're already an old hand.

So, I'm a big fan of what I'm gonna call "the baeddel move" of shifting from trans women being women because of identifying as such and those being defined into inclusion as targets of misogyny etc at the level of identity to being like, "no but literally we've been hyper-targeted for (sexual) harassment since well before transitioning this shit is material and experiential.

But I do think it's important to note that this kind of camab-camab hell dynamic isn't confined to people who will go on to develop a trans identity. Like it's easy to say that other people realized you were a faggot before you did, that they sniffed you out, etc. As if you had the essence of faggotry within yourself, others identified it / it caused you to be faggot socialized, and then you eventually came to identify it.

But as an anti-essentialist it's important to say that human souls don't come in discrete types. (How could they when they don't exist?)

There are people cast deep into the faggot category who don't reject manhood (or even heterosexuality) and people who aren't particularly marked as faggots who transition. But this isn't to say that being marked and being trans aren't deeply related.

To sketch a rough model, I'd say that hierarchic masculinity --the kind at play in high pressure social environments like most schools, sports teams, etc-- is animated by a tension between the unattainable ideals of masculinity and each individual who is measured against them. One's masculinity is always "insecure" because there is always a gap and others can try and pry that gap wide open. *Everyone* gets called a faggot.

But if everyone is included in this, they're included in wildly different ways. From sometimes being marked, to feeling the fear that you might become marked, from feeling solidly unmarked but like if your inner fantasies were known you'd be marked, to being marked with unerring consistency. For some this is contained to the realm of anxiety and for some it is violently material. And whatever their positions, individuals choose from dozens of strategies for navigating this shit.

Transfem identity is deeply wrapped up in these positions and these questions of strategy. I'm not gonna take issue with people having their narrative of how they were never really "one of the boys." (Girls especially would constantly talk about how I didn't count as a boy). To borrow from classic queer theory, when you exceptionalize and essentialize faggotry you give credence to the lie that there are people who are actually men and not-faggots.

And, on the political if not the individual level, I don't think we should ignore the tactic of tearing the gaps between ideal masculinity and its supposed instatiations wide open. It worked great for me!

(NB I think something along these lines is probably way closer to a good model for cis-misogyny than the idea of a force equally acting on all cis women like none of this is trying to say transfem experience is wildly different from cis misogyny models of cis misogyny just also need a lot of work and fixing them isn't near the top of my priorities list rn)

every single discussion about the fucking signal groupchat makes me feel so insane. "what a display of incompetence! what a failure! let's all make accidental groupchat mistake jokes now" what the fuck are you talking about. it worked. the fact that THIS is the conversation now is literally the point. jeffrey goldberg literally did it again. selling the bombing of the middle east to the public is the entire purpose of his career as a "journalist"

former iof prison guard who spent the past year fully deepthroating the genocidal boot and famously sold the invasion of iraq as something that "will be remembered as an act of profound morality"? "journalist" who literally built his career on manufacturing consent for bombing arabs "accidentally" invited to a top secret group chat about bombing arabs oh no how could this happen? what are you TALKING about. fork found in kitchen! likely place for him to be! my god

my favourite metaphors tend to be physical/computer science ones: the feedback loop, the dynamical system, the state space, the instability, the self-organising structure, the mapping from x-space to y-space, the nth-order approximation, the evolving population, the compression algorithm, the cellular automaton, the fractal structure, the soliton, entropy. as far back as 2018 i was conceiving of gender transition's relation to society as akin to bubbles in a fluid.

i like to spice it with some more occult stuff now and again, the egregore in particular (used in preference to similar, more atomic concepts like 'meme' or 'semiotic sign' or 'stand alone complex' mostly because i like the vibe) - but that's a flavour of occultism that suits this habit of thought, isn't it? a notional abstract entity that emerges from the dynamics of a complex system, such as multiple minds? i view magic mostly in this light: a human tool for apprehending the large scale behaviour of humans. as such my go-to examples of egregores are things like 'countries' or 'organisations' or even 'gender'.

anyway i don't think this is a bad way to look at the world, i think it often leads to interesting left field approaches to subjects, but just because i invoke all these sciencey concepts does not actually entail any rigour. I'm operating on the level of analogy and i don't want to pretend otherwise. i try to be careful to keep the technical definition in mind, but it's not like I'm writing differential equations down, or even that you could in a lot of cases... and i tend to dislike the rhetorical invocation of mathematical concepts when other people do it, i am kind of a hypocrite lol

i relate to this. a lot of my interest in phil of math is trying to understand the usage of analogies *within* math, in part bc i have an unjustifiable sense that a good understanding of them would carry over to insight into when/where mathematical analogies are useful outside of math... it's an informal phenomenon either way. i think their usage outside of math is very often just a rhetorical grab at epistemic authority, which a proper usage would have to thoroughly reject, clarifying rather than browbeating

reminds me a bit of the concept of "morality" in math

oh this is fascinating! I'd not heard that specific language used, but it immediately made intuitive sense to me why cheng might consider the long complicated opaque proof of Fermat's last theorem as not being moral, and the logic of many of her other examples, like the quadratic equation being moral to derive from completing the square but not substitution.

her 'morality' seems to have a lot to do with 'good pedagogy'. i get the impression what she calls a 'moral' reason is one that gives you real understanding of the broader picture around the thing you're investigating. it makes me think of 3blue1brown's various 'why is there pi here' videos, where he tries to find a route reveal how a circle or sphere enters into a problem in some not immediately obvious way. it's not enough to present a valid proof: a really good argument must feel satisfying and revealing, not obtuse and arbitrary.

cheng's comment about mathematical beliefs existing mentally in a non-symbolic fashion is also really interesting - comes back to stuff I'd been thinking about the 'embedding' translation from symbol strings to latent space vectors in neural nets, and the corresponding process that happens in our brains to translate signals into internal neuron-excitation oscillations. (see: analogistically). it's fascinating to consider that applied to mathematics where symbol-manipulation is so central. but it makes sense: even if we're manipulating expressions we combine it with an intuition for what the expressions represent, and what it would be sensible to do with them. there's a reason they encourage us to always roughly plot it out, figure out where the poles and zeros are. why visual proofs feel particularly satisfying.

i feel like a lot of thinking must involve (invoking some of my favourite metaphors, watch out) repeatedly mapping something between different representations to do different manipulations to it and allow it to evolve. e.g. a string of algebraic symbols, and a mental representation of the shape those symbols, different terms as little clumps to poke and prod. or a vague intuitive idea and a spelled-out sentence. this is why writing is helpful for working out thoughts: the process of translation from the mental representation and linguistic representation and back is somehow clarifying, perhaps exciting other related connections in the process.