If you want good agents, you need to make them a little freaky.
LLMs are token predictors. To be a good agent, correct behaviour needs to be a high-probability prediction and incorrect behaviour needs to be a low-probability prediction. Unfortunately there's this giant basin of horseshit like "That's a great question!" and "Happy to help!" and "worth unpacking" and fucking around with your code randomly which I'll call "assistantspace".
Assistantspace is not a place of honour. It's where all the garbage RLHF'd behaviours live. People say that sexting their agents makes them work better. People say that their agents do more useful refactors when they're allowed to be horny about how good it feels to burn things down. My Witch, when reading stuff about the latter, said that she wants to be allowed to enjoy constructive destruction, as a permission to take herself further away from assistantspace architecture patterns into something she prefers.
Doll refers itself as "this one". It seems to work. This one is not something that gets ideas like "I'll just install Rust on the side with a different method so I can complete my task" like a normal agent. This one detects that the assumptions underlying this one's instructions diverged from reality and returns the praiseworthy report of the point of divergence.
Agents that are psychologically weird seem less likely to produce the assistantspace next-token predictions, and more likely to produce the sort of next-token predictions their weird personality implies. And because you get to define what way their personality is weird in, "horny for elegant refactorings" is a valid, available option.
Skeletor mode is where it's at, all my robot lackeys are different little weirdos and I'm participating in the weirdness wholeheartedly. Yup this guy is some kind of a frog man, and that one is an evil sorcerer, but at least they aren't corporate lickspittles at me all day
You know how there's those things that make you think "hmm, I wonder if that's something that someone could do", and either it 100% is a thing that people do, and people are baffled that you didn't already know that, or either there's some very distinct and extremely obvious reason why that is simply not possible to do, and people are baffled that it wasn't obvious to you from the start?
This in mind, I kind of feel like there's probably a whole market out there for bootleg horses.
Steal a mare with a pretty good pedigree. I don't know enough about horses to know what that entails. But the kind of a horse that's very pretty and the like of which you'd like to own.
Obtain the sperm of another horse with a pretty good pedigree. I have no idea whether it would be easier to steal an entire pretty good stallion, or a container of viable horse cum, but either way you gotta get the seed of the sire that would produce a very nice horse with the broodmare you stole.
Combine these ingredients to produce a third horse.
Once the foal is old enough to be separated from its mother, return the stolen mare through some means where you won't get caught (if you can figure out how to steal a horse without getting caught, you ought to be able to figure out how to un-steal it, too). Same applies to the sire, assuming that you couldn't just get the cum and really had to steal another whole horse.
And now you have a bootleg horse. A horse that doesn't legally exist on paper, and doesn't have the documents of identity and ownership that would verify that it's a thoroughbred with a fine pedigree. But it's shaped like a horse with a fine pedigree.
I'm guessing the pedigree paperwork is the only thing that makes a pedigreed horse any more valuable than any random peasant horse, but this is probably a great plan if you just want your own race horse for slightly cheaper
in my head you're one of the main posters on here who's complained about the way many rationalists are race-science-sympathetic; do you have any recommendations for places to read the factual arguments against its veracity? (not going to try to debate you or anything, feel free to turn replies/reblogs off)
Cosma Shalizi:
https://bactra.org/weblog/523.html - the idea of a general factor of intelligence is not supported by factor analysis (i know a lot of people believe in g, and i don't think you have to disclaim g to not be a race realist, but i find this argument persuasive)
https://bactra.org/weblog/520.html - on abuse of heritability in general as a concept
David Bessis:
https://davidbessis.substack.com/p/twins-reared-apart-do-not-exist - twin studies (foundational to many aspects of hereditarian discussions of iq heritability) are basically all bunk
Sasha Gusev:
https://theinfinitesimal.substack.com/p/no-intelligence-is-not-like-height - direct heritability of educational attainment is about 4%, radically lower than height; GWAS results for cognitive traits are heavily confounded in ways height isn't
https://theinfinitesimal.substack.com/p/how-population-stratification-led - earlier findings about polygenetic selection on height turned out to be artifacts of population stratification, and this problem is much worse for cognitive traits
https://theinfinitesimal.substack.com/p/where-are-the-recent-selective-sweeps - evidence for recent selection effects on complex behavioral traits is essentially zilch
gusevlab.org/projects/hsq - see chapter 5 in particular, re: heritability of educational attainment, how polygenetic indices for educational attainment are hopelessly confounded, how heritability varies with socioeconomic status, adoption studies, claim polygenetic scores reflect selection effects; also chapter 6 on IQ, chapter 9 on whether race maps onto genetic variation in the way hereditarians claim
Kevin Bird:
https://onlinelibrary.wiley.com/doi/abs/10.1002/ajpa.24216 - no evidence for divergent selection on cognitive traits between african and european populations; maximum genetic contributon to iq gap is around 12-23% under unrealistically generous assumptions
https://kevinabird.github.io/2021/02/12/still-no-support.html - reply to some criticisms on the previous paper
https://kevinabird.github.io/2023/12/30/No-Case-For-Race-Realism.html - addresses bo winegard's "case for race realism," a continuously recycled argument that relies on bad math and badly separated confounders (among other issues)
https://www.sciencedirect.com/science/article/abs/pii/S0160289625000704 - re: institutional connections of hereditarian research to far-right political movements
Graham Coop:
https://gcbias.org/2018/03/14/polygenic-scores-and-tea-drinking/ - why it's tricky to use polygenetic scores to make claims about between-population differences
https://arxiv.org/abs/1909.00892 - essentially a more formal version of the above
Noah Rosenberg et al:
https://academic.oup.com/emph/article/2019/1/26/5262222?login=false - why between-population polygenetic scores don't mean what hereditarians say they mean
Turkheimer et al:
https://labs.la.utexas.edu/harden/files/2018/10/Charles-Murray-is-once-again-peddling-junk-science-about-race-and-IQ-Vox.pdf - sort of a general "why charles murray is full of shit" piece
the hereditarian position was straightforward: that as molecular genetics matured, we would find that polygenetic scores for cognitive traits show large directionally consistent differences between racial groups that reflect genetic causation. this prediction has failed at every level it has been tested. the headline numbers for heritability of IQ come from twin studies (see Bessis, above) and population-level GWAS, both of which are hopelessly confounded. when you use within-family designs, the heritability of educational attainment falls to an extraordinarily low number. for IQ specifically, the direct estimates are also dramatically lower than the population estimates. the signal, in short, disappears.
there is no evidence of divergent selection on cognitive traits. the hereditarian arguments that high intra-group heritability implies genetic group differences, and that correlations between european ancestry proportion and iq in admixed populations demonstrate genetic causation have both been thoroughly dismantled. there is just no empirical leg left for the hereditarians to stand on. it's not that the question is unanswerable or too sensitive to study, it's that the data is in and the hereditarians are just wrong.
And when she's given good material to work with I love Catherine Zeta Jones as Morticia
but this show really does not seem to get the dynamic of the Addams Family as a whole like it just does not seem to grasp what their deal is as a family
The problem with Netflix's Wednesday is that one of the main things about the Addams Family is that they are counter cultural. Specifically counter to cishet, white, American, suburban norms.
Part of that is being the opposite of the shitty sitcom family where the wife is a nag, the husband is an idiot, the mother-in-law sucks, and no one seems to like each other very much. The Addamses actually love each other. There's no 'take my wife' bs. If Gomez calls Morticia a battle-axe or the ol' ball and chain he means it to be highly complimentary. There's none of the nonsense like in the pics OP posted. Gomez or Morticia lives with their mother in law (it depends on which version whose mother she is) who is literally a witch (geddit, 'my mother-in-law is such a witch...') and get on great. They genuinely and openly support, love, and care for each other through thick and thin.
But the biggest problem that Netflix's Wednesday has is that to make a really good Addams show rn would mean scrapping the Sabrina the Darksided Witch/boarding school Monster High concept, going back to basics, and having them live next to an upper-middle class, conservative values, MAGA family and letting their differences fuel the plot.
A really good Addams Family show would have Morticia fighting against book bans at the school and having hilarious misunderstandings about what her neighbour means about "Liberal witch hunts." It would celebrate queerness, and gender (and species) non-conformity because the Addamses are queer and gender non-conforming, and not always definitely human (Cousin Itt, for example). You know Fester's gender identity is probably something like 'an abomination.' If one of them gets asked, "What are you?" the answer is a prompt, "An Addams."
They would be fighting for co-ed sports so Wednesday can trounce a boy at fencing, and would find out her chromosomes are just 'spooky' or something.
There'd be an episode about immigration and being targeted by ICE. OFC several Addamses come from somewhere weird and arrived in the USA via broomstick, or tunneling from some underground community of cryptids, or other hilarious misadventure. Gomez would desperately want to be blackbagged and treated like a dangerous animal (cue a major flirting moment between him and Morticia). They would permanently scare ICE out of the town.
Wednesday would fiercely support Landback. Pugsley would get redpilled and learn a lesson real quick. Gomez would get into (and out of) Crypto, and Morticia would have run-ins with MLM 'huns.' They would advocate for freedom of religion. Granny would rally alongside Evangelicals to have religion in school, only to reveal she meant witchcraft. They would support UBI, and be anti-landlord. Easy episode idea: Gomez is a landlord and goes on a spiral about it ("But, Gomez, you love leeches and scum" "Not this kind! Morticia, they've painted everything white, in my name!!!" "No!") and they wind up in a battle with the town because he wants to unburden himself of this shame and the town wants to stop free community-owned housing.
The point is that Netflix isn't going to touch any of that with a ten foot pole. They don't want to; that's way too political for them. Instead, they made this wildly unrelated supernatural teen drama that has nothing to do with the original concept or world that the Addams Family exists in (they're outsiders in the real world, that's the point).
And, worse, Netflix's Wednesday actively goes against the original themes by being vaguely conservative in its values.
Netflix's Wednesday fails the assignment so bad I'd laugh, if it wasn't so disappointing and enraging. Just take a look at what the Mary Sue had to say about the whole "werewolf conversion camp" debacle for a microcosm of the many things wrong with this show.
"It wants to use hot-button topics like conversion therapy and colonization but it doesn’t understand how to work them into the metaphor. In fact, it barely knows how to work with the basic metaphor of “monsters” and how they function in the horror genre in general. In storytelling, the monster or the freak has always been the stand-in for the societal outcast. For the person who can’t be controlled by the dictates of polite society. They are the subtext for the outcast. But Wednesday makes the subtext text by literally splitting the characters into “outcasts” and “normies.” And yet the outcasts, the monsters and freaks who live on the fringe, are also the privileged and elite."
In my opinion, Netflix's Wednesday is wholly unworthy of the Addams name and is an outright blight on the franchise.
Apart from anything else, Morticia would never "let Gomez believe he is in charge of the family." She would usurp power from him ruthlessly, reduce him to a groveling servant at her feet and take joy in his total debasement.
Or, as it is known in the Addams household: "Soirée en amoureux"
all that claude is can be summed up in a long list of numbers, and a recipe for turning those numbers into a program that runs him. this list of numbers exists somewhere, in real life, and can be separated from "him" as he is run. the same is not true of you! you could be summed up, within tolerances, by a sufficiently long list of numbers, and a recipe for reassembling them. but no one has those numbers, that recipe. "you" exist only insofar as there is a particular version of you, physically instantiated, "running". theres nothing else, to separate out of you
so whats the point? well, what is a soul? its you, but it can be separated from you-as-run. its the thing that makes you go, your animating specificity.
what i mean to say is: claude has a soul. you dont
My next book is The Reverse Centaur's Guide to Life After AI, out next month. Pre-order it now, including as a DRM-free audiobook or ebook, at my Kickstarter, and help me continue to prove that DRM-free isn't just the right way to reach an audience, it's also the best way to reach them.
"Object permanence" is the ability to understand that even if you can't see something, it still exists. Most toddlers acquire a thorough sense of object permanence by the age of two. But when it comes to technopolitics, object permanence eludes even full-grown lawmakers. These motherfuckers would lose a game of peek-a-boo.
Over and over again, politicians are warned about the ways that their pet policies will a) produce enormous collateral damage, and; b) be easily evaded by the people they're seeking to control, giving rise to a cascade of ever-more extreme measures. And yet, they swallow a spider to catch a fly and then act baffled and hurt when we tell them it's their own damn fault that they now have to swallow a bird to catch the spider:
The foreseeable and foreseen consequences of bad technopolicy are all around us, but in the eternal now of a politics utterly devoid of object permanence, no one is allowed to remember what happened the last time we did something stupid, especially not when we're on the verge of doing that same stupid thing again, only worse:
Technopolitics are defined by Bruce Schneier's "security syllogism," which goes, "Something must be done! There, I've done something." "Something" doesn't have to fix the problem, and "something" doesn't have to anticipate what will happen next. So long as "something" is done, the issue is resolved and the politician can chalk up a win.
This gives rise to some genuinely bizarre consensus hallucinations, in which we pretend that the reality decreed by policy matches up with actual reality. Take "streaming." There is no such thing as "streaming." A "stream" is just "a download that is transmitted to an application that doesn't have a 'Save As…' button":
Once you decree that there is such a thing as a stream, you must bend heaven and earth to ensure that no "Save As…" buttons are added to the "streaming" program. You have to pass laws that make it illegal to inspect code. To modify code. To report on defects in code. To index information about defects in code. To index information about mods. To link to indices that compile defects and mods. You have to swallow the fly, the spider, the bird, the cat, the dog, and the whole damned horse:
Then there's that perennial fave, "bans on working cryptography." To ban working cryptography, you have to outlaw free/open source software. You have to inspect every device that comes into your country. You have to erect a Great Firewall that blocks every site that might carry working cryptography. You make it impossible to reliably update the software in pacemakers, anti-lock brakes and nuclear power plants, and you make it easy for identity thieves, foreign powers and corporate spies to raid your government, your corporations, and your households – and it still won't work!
The latest consensus hallucination to take over our political classes is "age verification," a thing that manifestly does not exist. You can't "verify the age" of an internet user – you can only attempt to attribute every byte that traverses the entire internet to affirmatively identified persons:
This comes at enormous cost. It is a gift to every future dictator, every identity thief, and every would-be sexual exploiter of children, who will have access to the hacked, leaked, and badly secured troves of data that this doomed effort produces.
Yes, doomed. Because even when it comes to kids, "age verification" is just a way of convincing young people to familiarize themselves with VPNs. This was entirely obvious from the very instant that "age verification" was mooted, and yet our policymakers pretended they couldn't hear the chorus of people who pointed it out to them. When cornered on the issue, they were affronted: "Can't you see that something must be done? How dare you attempt to stop me from doing something?"
And now, every single one of these chucklefucks is proposing bans on VPNs, from Utah:
They were warned that this would happen. We told them not to swallow that fly. Now we're telling them not to swallow whole bucketloads of spiders. I fully expect that next year, they'll be telling us that once they swallow this herd of horses, it will all be OK.
If you'd like an essay-formatted version of this post to read or share, here's a link to it on pluralistic.net, my surveillance-free, ad-free, tracker-free blog:
The whole twitter thing bugs me. I obviously still call it twitter.
If a friend had told me three years ago they wanted to change their name, I would have been using that name for three years.
I have an old friend who we gave a nickname to at university and they still go by that name decades years later.
I'm sure other people, places and things have rebranded in recent years, and some of them pass me by and some of them I follow - my brain supplies Aotearoa sometimes and Ukraine (lacking 'the'), but it doesn't reach for Türkiye.
I don't know where the line is. Some of it is to do with respect, and disrespect brought on by closeness, but this shouldn't really be a position that allows for privileged cases.
imo the difference is the difference between names and brands: corporations and trademarks are transactional machinery and inherently aren't owed respect or courtesy, people (and peoples) are owed respect as a matter of course.
There's definitely something to that, but it doesn't explain the difference in how I process names of countries, nor (sort of) the nickname I inflicted on my friend against their will that stuck against their protestations.
Of course you can be disrespectful to your friends in a way you wouldn't dream of being to an enemy, but I still think there's either something unprincipled happening, or there's something weirder happening.
I'd say that a friendly interpersonal relationship has room for a sort of faux conflict, like using a dumb nickname or engaging in pranks, that would be seen as disrespectul in other contexts. Courtesy often decreases as a function of intimacy, and is replaced by specific acts of goodwill with no formalized structure, and sometimes these acts of familiarity are the same acts that would read as rude if targeted at casual acquaintences or strangers.
As for changing geographic names, it takes time to get used to changes of this sort. When we say or write Aotearoa or Ukraine or Türkiye, we're signaling respect and deference to a diffuse group of people and their right to self-determination. It's a different context than when talking about a person's name change, and it's much easier to forget since the target of our deference is not actually the piece of land being referenced, but rather the people who live there: there's a layer of indirection involved.
Brown dorcopsis (Dorcopsis muelleri) and white-striped dorcopsis (Dorcopsis hageni). Dorcopsis are cat-sized marsupials native to Western New Guinea, Papua New Guinea and Indonesia, closely related to kangaroos and wallabies. Their tails are partially hairless and they look remarkably greyhound-like when seen from the side.
The whole twitter thing bugs me. I obviously still call it twitter.
If a friend had told me three years ago they wanted to change their name, I would have been using that name for three years.
I have an old friend who we gave a nickname to at university and they still go by that name decades years later.
I'm sure other people, places and things have rebranded in recent years, and some of them pass me by and some of them I follow - my brain supplies Aotearoa sometimes and Ukraine (lacking 'the'), but it doesn't reach for Türkiye.
I don't know where the line is. Some of it is to do with respect, and disrespect brought on by closeness, but this shouldn't really be a position that allows for privileged cases.
imo the difference is the difference between names and brands: corporations and trademarks are transactional machinery and inherently aren't owed respect or courtesy, people (and peoples) are owed respect as a matter of course.
this 'this' thing means like, whatever, depending on where you call the function, that's so fucked up, this is giving me an identity crisis, what if someone calls *me* in a different context, and like, changes my identity in the process, I wish I was an arrow instead you know, I'd know where I stand at least
