senior fellow at the Manhattan Institute, former chief economist to Senator Rob Portman (R-OH) & director of budget and spending policy for Marco Rubio’s 2016 presidential campaign👇🏽
Social Justice Bard. They/them. Gay; tired. Joy is a strategy. "Where does the Constitution live? On paper? Or in our hearts?"
2,741 followers 2,684 following 8,011 posts
view profile on Bluesky
senior fellow at the Manhattan Institute, former chief economist to Senator Rob Portman (R-OH) & director of budget and spending policy for Marco Rubio’s 2016 presidential campaign👇🏽
Using one photo already on your phone, what is the announcement tomorrow at 2?
A lot of posts on this site, including posts that are very well received and made by very smart people who *can* do better, on a variety of topics, are recapitulation of slogans. Regurgitation, you might call it
I taught my last class about 7 years ago and the average amount of thought put into an essay submission was not much higher than the amount of thought put into a one paragraph AI prompt. Most people are not reasoning most of the time. So much human activity is autopilot
The other thing about AI discourse that really gets to me is that I absolutely give a lot of leeway to anti-AI people who work in education because, again, I used to teach writing, but a lot of people who haven’t been instructors have incredibly optimistic/sanguine ideas about human reasoning
This piece is very important on its own merits and resonates with some initiatives I’ve seen for other causes like sex worker rights, reproductive justice, & harm reduction. You can’t just show up and ignore stakeholders/communities that your idea is more solid than their experiential knowledge.
wrong temperature and time of day but generally a real solid option my brain eventually settled on "sparkling water with a bit of rose syrup added" and it is delicious
ugh i legit wish i had some non-alcoholic beer. or like a REALLY big can of cheap shitty american pisswater (positive, complimentary) pilsner. i have successfully noticed Thirsty, but i Do Not Desire Water and I'm like "flavored, sparkling water? :)" and my brain is like "go fuck yourself, buddy!"
ANYWAYS IT'S SURE LUCKY THAT A BELOVED AND CRUCIAL POLITICAL COMMENTATOR WITH WHOM WE ALL STRAIGHTFORWARDLY AND COMFORTABLY AGREE HAS BEEN MAKING THIS ARGUMENT FOR YEARS ALREADY AND HIS NAME IS WILLIAM STANCIL [falls into an overstuffed armchair, sobbing]
lmaoooooooooooooooooooooo
*stares into the distance for several minutes*
i will buy your eggs but i will not become your new chicken and it is unreasonable that this has become the foundation of so many goddamned interactions
and the thing is, i recognise this pattern at homeopathic dilution, it is bullying. it is a form of social power. i can fully agree with a specific salient claim that they are making, concede that point, and then that is not actually what they meant or where they want to go with it
no, I am not going to give examples, because I love myself too much for that, although many of them are political and if you've followed me for any length of time you can make certain inferences, and your inferences will probably be correct
there is a set of Discourse Topics which you can go absolutely out of your way to Perform the Proper Obeisances and precommit to saying "baby don't worry baby i know you're right baby" every other sentence but still if you try to engage with nuance and clarity the wheels fall off
I HAVE HAD THAT. it was INCREDIBLE lol. I did not expect it to work, at all, and it somehow did
huh I'm kind of surprised I haven't run into that yet. I haven't had a Southern Tier I didn't enjoy, and they've pulled off some absolutely wild nonsense
lmaoooooo yes continued adventures of Data being the character I vibe with the most I guess
Earl Grey tea, that's a *name*, you have to get that right human hair regardless of its place on the body is grey, i think because humans are more like cats than like the sky, but humans are also like the sky in important ways so they can have gray hair, it's just spookier
both grey and gray are correct but in my head i have to use the RIGHT one for the right word like skies are gray but cats are grey
both grey and gray are correct but in my head i have to use the RIGHT one for the right word like skies are gray but cats are grey
I don’t think Kamala Harris would have pushed for the U.S. to literally annex Gaza and build a resort town on the rubble
somehow I picked up British spellings for words around the same time I drowned my natural accent in the middle of the Atlantic Ocean (I think it's actually that I grew up on the Internet with a lot of international and UK friends and was a rebellious teenager during the Bush administration)
But they *are genuinely engaging in symbolic transformation characteristic of linguistic reasoning*, probably through methods entirely distinct from how human brains do it This is just a thing machines can do now. This has been an open secret longer than the marketing hype
They also are not engaging in the full range of cognition or reasoning that humans express, because, again, they’re just language models, they have nothing to work with except language itself, they do not have senses or hormones or any of the dozens of other perceptual faculties
They aren’t sentient, because sentience is a specific thing defined in relationship to biology and involves systems that can be observed in nature that are not present in these models, and which the models expressly aren’t doing (you can also observe this just from using them +looking @ their arch.)
But a lot of laypeople absolutely have allowed themselves to become convinced that transformer models are doing something they in fact are *not* - rearranging or collaging the training data, which is impossible and you can prove it by looking at open weight models
The thing that is irresponsible that these companies are doing is that they are selling a product they do not understand the functioning of fully, and that they are asserting that it is fit for purposes they have no non-motivated reason to believe it is safe or fit for
We still do not fully understand how exactly a bunch of token weights interpreted by an inference algorithm can produce the kinds of connections between symbols and constructs that these models are producing, i.e. why they mimic thought so well That part is *not marketing hype*
There’s another point here I forgot to make and want to finish elaborating: we still do not fully understand the reasons why these models work at all, and all the “training for the purpose of being a chatbot assistant” stuff is just stuff they’ve discovered works through, basically, trial & error.
If you have an internal combustion engine, you do not *have* to make a car there are things completely unrelated to cars these engines can work in. and there can be cars without internal combustion engines. same thing applies to chatbots/LLMs.
To put it another way: we can imagine a society without cars but with an internal combustion engine. We can argue and debate whether this would have emerged, and we obviously live in a world where the car *did* emerge from ICE engines. but they are not the same thing.
We're dealing with a genuinely novel and interesting behavior emergent from an algorithm trained on large corpuses of human communication, that has been *specifically adapted with a specific cultural and economic goal* to a specific purpose which is, in many ways, socially delitirious
But the thing is that a chatbot is *not* the only conceivable expression of "language model trained on an underlying corpus of text," and you can see this from coding models and how a lot of the way they're used isn't that much like a chatbot interface at all
I think AI coding assistants would probably still be taking off; we probably wouldn't be calling them AI but "generative coding" or something like that; and AI image generation would probably still be out there but not being commercialized or promoted as much, mostly a hobbyist thing
I hypothesize that if the end of the zero-interest-rate policy era hadn't led to a funding cliff in the early part of this decade, causing OpenAI to pivot to creating and releasing ChatGPT as a product, there would still not be a large number of highly commercialized chatbots out there
Gemini and Claude are less like this because both Anthropic and Google have somewhat higher-brow aspirations for the product; they tune the models for scientists and coders and envision them as something like research assistants, and you can see this in the different behaviors they manifest.
OpenAI has trained ChatGPT (and their open source GPT-oss models) to be, as far as I can tell, affirming, supportive, and friend-like - which I consider to be *colossally* unethical, particularly in combination with the other marketing they've done around it it's actively deceptive to the user
most of the "product" if you use ChatGPT vs. Grok vs. Gemini vs. Claude (or whatever) is in these conversations that, basically, tell the model what it is for and what its *goal* is, since its training data is made up of a bunch of stuff that is *not* "conversations w/ a helpful AI assistant"
the other day I did that thread about how I noticed OpenAI's models were overwhelmingly less receptive to a prompt I created to try to address the sycophancy &user-indulgence issues many folks have noticed this is a product of how OpenAI does their training to the unbaked model,vs.other companies
this training (which, in my opinion, is in many ways at least as ethically dubious as scraping the web or training on intellectual property, bc these are *not* well-paid engineers in most cases) involves training the chatbots to behave like machine assistants for particular purposes
this isn't *inherently* sinister, or at least I don't think it is the common way that these models are tuned beyond the inclusion of the (still overwhelmingly organic and human sourced) training data (from web scraping, books, articles, etc)is through injection of interaction data w/human employees
*now that AI models have been released as commercial products*,they are being tuned to serve specific commercial purposes,and because those commercial purposes are in many cases things that their creators/publishers are trying to create to get money,they're being tuned to persuade us in various ways
but those early, "unbaked" language models were *basically* produced by applying math to various subsets of human generated data about humans. nobody was "designing them to fool people" because they weren't for public release or consumption, they were pure research products
when I say "genuine novel spooky stuff" here,I mean that it is still not at all clear where a lot of the emergent behavior in these models comes from,there's a lot of active research into it and,crucially,the fact that a lot of this research is being done by corporations w/vested interests clouds it
1) the training data that these very early (pre-GPT3) language models worked from was comprised entirely of human communications, so the models claim to be human because their source data is all within the reference frame of humans communicating with other humans; 2) genuine novel spooky stuff
we can, again, get into all kinds of speculation about "are people who work on AI predisposed to believe they're creating superintelligence" (which I don't think they inherently are, but I can see why you'd suspect otherwise) but the reason these models did this is bc of at least two things:
but, to disambiguate for a second: people at Google and other companies were starting to experience what we're now calling AI psychosis by the end of the previous decade, from interacting with raw, untuned language models turned into rudimentary chatbots. these were not designed to fool anyone.
this is something I don't think it's at all unreasonable to be unclear on because if you don't have a really solid understanding of how the different elements of these models interact, it'd be very easy to confuse different parts of the process that got us to "ChatGPT driving people to suicide"
I think anti-AI people are also predisposed to believe what they want -specifically, that these models must be being designed to "trick" people, and the commercialization and tuning of them is giving people who don't get deep into the technical reeds cover to believe this bc it's got grains of truth
This is an area of research where we absolutely are not going to get any good answers as to how it works and why it’s able to do this without being sentient if we remain committed to the erroneous idea that there’s nothing impressive or novel about it, or that it’s simple tricks
so, again, with the acknowledgement that there have been people working in the field who went in expecting to create some kind of superhuman intelligence and that those people are predisposed to see what they want to see, it's also true that it is legitimately fascinating what transformers can do
I think another thing that is insufficiently understood is that it is both true that there are people working in the tech industry who are susceptible/predisposed towards quasi-religious thinking about this stuff and that also the technology works in legitimately uncanny and unexpected ways
what happened was that they figured out that you could get spooky-good predictive text with a bunch of fascinating emergent properties, and that similar kinds of attention-based modeling could be applied to visual and auditory artifacts, and they needed money to keep researching that
there are people who've been trying for the past few decades to develop something they imagined might turn into "AGI", but that was originally roughly like how some political scientists imagine their research might eventually lead to full communism - it didn't mean that was the goal of the field
attention-based transformer models were not developed with the goal of "we want to make an AI therapist" or "we want to make an AI girlfriend" or "we want to figure out how to supplant white-collar professionals" all of those things are attempts to find a market application for an algorithm
what you are seeing, with AI being forced into all corners of life by marketing, is an effort to generate revenue and investment interest around a specific set of discoveries in computer science that do not inherently suggest particular applications or purposes
I just wanna grab this part of my ramble about AI from all of this; I think it's insufficiently understood the extent to which AI models as they are currently being commercially deployed are an effort to commercialize a specific novel technological discovery - the transformer - for funding
there is also Startlingly Kinky Adam Savidan Content please look forward to it
holy shit Cameron Lauder lmao the Loading Ready Live cold open
CHICAGO IS IN THE STREETS! #LaborDay #DefendHigherEd #WorkersOverBillionaires
Together, we've made Illinois a beacon of hope for workers everywhere. We've made life more affordable for families, created good jobs with fair wages, and made it easier to get an education and find a career. When we stand up and stand together, we win.
Hi, what the fuck is going on at Walter Reed?
hell yeah hell yeah
my brain was like "well the eyes suck and the ears may as well not be there. what do we have on the ship that's GOOD" and in another life i'm designing perfumes about it or something, a thing i actually really wanted to do as a kid, & i made perfumes for family members for a while, I was good at it
i think genuinely all of my ability to perceive fine detail went into wringing as much out of the senses of smell and touch as the human body in its artless and undignified limited fashion will permit me
i have an interesting relationship with flavour
a cocktail i deeply love is one where, every time i have the first sip of it i am like "this is poison, and i am going to die", and then the second sip is "brother why did you take a second sip of poison?", then the third is "...huh" and after that it's delicious this happens EVERY TIME
... which is hilarious to me because right-wing 501(c)(4) groups violate that law *for a living.*
Unfortunately the author of the piece just... doesn't understand what "dark money" actually means, or how the laws around campaign finance work. So she heard a telephone-version of that restriction and turned it into a "big exposé." This is why we can't have nice things lolsob
We can still endorse whatever candidates we want. We just have to produce endorsement videos/graphics with other funds. (Like Patreon, product sponsorships, etc).
There's one (1) restriction: "you can't use grant $ to make content that endorses a candidate bc that's illegal to do with 501(c)4 money." It doesn't matter if the candidate is progressive, centrist, or whatever. That's literally a "keep dark money out of politics" law that everyone has to follow.
The scary headlines about "Chorus, a sEcReT dArK mOnEy gRoUp" have been absolutely wild to watch. Because I'm in Chorus, & it's been a godsend. Without it, I wouldn't be able to cover the US's massive modern-day farm slavery problem- at all. youtu.be/f5UwwtrGdak
I don't disagree with the first two sentences, but I think in general there's been far too much movement from 'not really noticing the problem' to 'despair'. Let's experiment with 'people and organisations try and enforce the norm' and then see if it is actually hard for it to be re-bottled.
but everything that does not descend to such abyssal depths -usually has something interesting going on worth exploring-; if it's just mundanely bad it'll be more BORING than anything else and I'll just set it down, it's like a reverse uncanny valley effect
side note that this entire excursion has been an exercise in why when my first reaction to something is "WOW THAT'S CURSED???" i always go back to it to try more to see if the flavor evolves. there is a dimensionality of malflavour beyond cursédness inhabited by CPU thermal paste and bad sea urchin
but everything that does not descend to such abyssal depths -usually has something interesting going on worth exploring-; if it's just mundanely bad it'll be more BORING than anything else and I'll just set it down, it's like a reverse uncanny valley effect
side note that this entire excursion has been an exercise in why when my first reaction to something is "WOW THAT'S CURSED???" i always go back to it to try more to see if the flavor evolves. there is a dimensionality of malflavour beyond cursédness inhabited by CPU thermal paste and bad sea urchin
i've had various of coconut candy and loved all of it, something about it in a *beverage* fucks with me. I cook with shredded coconut and coconut milk/cream a fair amount.
yeah the primary flavour component in cilantro and coriander is a scent that's VERY strongly associated with cleaning products in the modern West and I'm *convinced* that's what's behind cilantro aversion b/c It Tastes That Way To Everyone
i wonder if there's a cilantro thing happening where because I did not grow up really strongly associating coconut with food and drink but DID strongly associate it with sunscreen, my brain smells coconut qua coconut and goes "ew! yuck!" the way people smell the limonene in cilantro and go "soap"
(the cilantro thing isn't a genetics thing after all - there's a gene that makes cilantro BITTER but the main thing people are reacting to is cultural; were you raised in a culture where limonene is in foodstuff, or in a cleaning agent?)
i wonder if there's a cilantro thing happening where because I did not grow up really strongly associating coconut with food and drink but DID strongly associate it with sunscreen, my brain smells coconut qua coconut and goes "ew! yuck!" the way people smell the limonene in cilantro and go "soap"
this beer actually breaks the threshold of "this was a fascinating experience and I do not regret having it" in the end accounting, I'm not sure it's *successful* at what it's trying to do but it's making a very bold statement, as opposed to JUST smelling like sunscreen
my best friend literally said to me this weekend, out loud, "I'm going to go on choosing to believe it, as a little treat"
NEW: This weekend in Trump's effort to make fascism happen. The Trump administration tried to illegally send children to Guatemala as troops walked aimlessly around Dupont Circle. This evening, at Law Dork:
lol at the idea of sending pictures of bacon to someone who doesn't eat bacon for religious reasons to, like, trigger them or something like oh no, you have made me drool, and experience the *extremely specific emotion* that's why I choose to constrain what I buy for religious reasons. oh no. oh no
i am reliably informed that there are coconut-related drinks which are good. i am willing to believe that this is true or take the statement on faith. i have yet to experience a good coconut-themed beer.
if i do breathe in as I take a sip I get this big smack to the face of hair product, though, and that's unfortunate. maybe drinking this one from the can WAS the right play and I fucked up
this is the CLOSEST a beer has ever come to using coconut and being good and i'm intrigued by what they're doing right. they're not afraid to let it be... murky. the mouthfeel is unbeerlike in a way that i find profoundly interesting. if i do not breathe in until it is in my mouth i... like it?
i poured it into my beer stein because it's not a shitty American pilsner (positive, complimentary) and I'm not a barbarian, and I'm [gestures vaguely in growing up poor] about wasting food and its vileness is oddly compelling, so I'm probably going to finish it i'm not happy about it, though
Michelle Obama: "please I'm begging you feed kids just one (1) vegetable" RFK: "some children deserve to die. there's a healthy way to eat deep fried burgers" Media: "these are the same, to me"
(this last statement is a joke, i expressly do not believe food crimes can be real. but oh my god people keep making coconut beer and it's. why. why. i guess this is what i get for unironically enjoying IPAs)
just had the most cursed beer i've ever had in my life it tasted like fucking goddamned hair conditioner, there was sediment at the bottom of the can coconut should not be an ingredient in beer and people keep making coconut an ingredient in beer and it should be Illegal Food Crimes to do that
"there is always an ideal solution that convieniently asks nothing of me and I will get angry if you suggest otherwise" has really been a driving ideological force in our politics since 2020.
Yeah. I was estranged from my mom fairly significantly at the bitter end, and not involved in her care in any significant way, but that was how things went with her, too.
ironic that a senator from so far out in the pacific is primarily interested in representing The Atlantic
Americans Astronomers watching watching US politics Betelgeuse 🤝 “It could happen at any time”
an old credit card-like object that you don't need any more and can readily consign to the deepest hell as soon as you're done seating the heatsink is your friend here, only touch the Vile Substance with the credit card