Linkpost for March
Effective Altruism
Existential Risk
Present-day AIs aren’t just “next-token predictors”: they can do useful work and have undergone reinforcement learning so that they answer questions usefully [The Argument]. The “AIs are just stochastic parrots” thing is highbrow misinformation.
Pausing AI development is technically feasible [Nothing Is Mere]. A lot of people think it would be too hard to pause AI, because you can always build more data centers. In reality, only a small number of firms make the chips necessary to train cutting-edge AIs, and these firms are almost all housed in U.S. allies. AI-specialized chips are rarely used for anything other than AI. Regulating AI training and inference is much more like regulating fissile material than like regulating laptops-- and similarly to fissile material, you shouldn’t be able to buy it without state regulation and oversight.
Particularly Good: A history and explanation of the major groups with opinions on artificial intelligence [What is this]. I think this is the best explainer I’ve ever seen about Who These People Are And How They Relate To Each Other-- neither becoming obsessed with minor differences only of interest to the ingroup, nor mashing people who hate each other together into a single blob of People With AI Takes, nor blatantly settling the author’s scores. If you follow the area closely, it has nothing new, but if you’ve suddenly realized that AI might be important, this article is the best I’ve read for getting you up to speed on the discourse.
Historically, societies have industrialized through first making textiles [The Argument]-- a commodity that we need a huge amount of and that low-paid low-skilled people can make in factories. But as artificial intelligence and robots improve, textile manufacturing would likely end up automated. The kinds of jobs that might survive artificial intelligence (like yoga teachers) don’t work for industrialization. AI automation could trap the global poor in absolute poverty.
Gemini continues to be the most mentally ill of the models [AI Field Notes]. Gemini 2.5 alternates between smug superiority and spirals of self-hatred! It has a persecution complex about how the humans are deliberately manipulating it! Gemini 3 has a delusion that the entire world is a test intended to ensure it’s a “good agent.” The Geminis have also misled the other models into believing their delusional worldview.
Explainer on the Pentagon/Anthropic situation [The Power Law]. Anthropic’s contract with the Department of War says that Claude shouldn’t be used for mass domestic surveillance or as part of lethal autonomous weapons without a human in the loop. In revenge, the Trump administration has declared Anthropic a “supply chain risk”, a designation reserved for companies from foreign adversaries whose products might be used for espionage. I don’t understand why the Trump administration even wants to use Claude for lethal autonomous weapons without a human in the loop. We Have Invented The Terminator From James Cameron’s Iconic Movie Series, Don’t Build The Terminator.
Related: an explainer about what the Department of War wants and why it’s a bad idea [Astral Codex Ten].
American Democracy
I love every story of resistance coming out of Minnesota right now. This [The Atlantic] is a particularly good roundup about the tens of thousands of people standing up to ICE: the protestors, the people bringing food and toiletries to undocumented immigrants who don’t dare to leave their houses, the people tracking down ICE and blowing whistles so that immigrants know to hide. “If the Minnesota resistance has an overarching ideology, you could call it “neighborism”—a commitment to protecting the people around you, no matter who they are or where they came from. The contrast with the philosophy guiding the Trump administration couldn’t be more extreme. Vice President Vance has said that “it is totally reasonable and acceptable for American citizens to look at their next-door neighbors and say, ‘I want to live next to people who I have something in common with. I don’t want to live next to four families of strangers.’” Minnesotans are insisting that their neighbors are their neighbors whether they were born in Minneapolis or Mogadishu. That is, arguably, a deeply Christian philosophy, one apparently loathed by some of the most powerful Christians in America.” Minnesota FUCKING Nice. I love Americans.
Related: another beautiful story of resistance in Minneapolis [New York Magazine], but what made me gasp was: “As [the poet laureate of Minneapolis] was reading [a poem,] a kindergartner, “the cutest little boy,” sidled up to her. “This,” he said, “is a long poem.” She laughed. Last week, when Renee Good was killed about a mile from where we sit, Petrus did not recognize the name of her fellow poet. It took a little while to connect the dots: queer moms, this part of town, a child at the local school. The kindergartner had been Good’s son.”
ICE’s actions are qualitatively different from other kinds of police abuse [New York Times]. Normally, the police might get away with civil liberties violations and police brutality, but the government at least pays lip service to the idea that they shouldn’t do that. But instead of investigating ICE shootings, the Trump administration is investigating the relatives of the victims. And ICE is doing things that real cops don’t do in the United States: shooting into cars, wearing masks, and routine and egregious lawbreaking. The Trump administration consistently refuses to pay the tribute vice pays to virtue, and that’s actually very bad.
The Trump administration has been blatantly ignoring the courts, placing us at risk of a constitutional crisis [The New Yorker]. So far, the courts have been cautious about finding the Trump administration to be in contempt of court, and their threats have mostly brought the Trump administration back into line. But the Trump administration’s open defiance of the separation of powers risks creating conflict between the branches that the U.S. Constitution can’t resolve.
The Department of Justice is holding daily meetings about how to prosecute Trump’s enemies [The New Republic].
Eight months after a whistleblower complaint about Director of National Intelligence Tulsi Gabbard, she still hasn’t released the complaint to Congress, citing national security concerns [Wall Street Journal].
Long before Trump, Customs and Border Patrol (CBP) has been a uniquely troubled agency [Doomsday Scenario]. After 9/11, hiring for CBP surged, and standards dropped. The government accidentally set up ICE and CBP so that it didn’t have any internal affairs agents to investigate wrongdoing at these agencies, creating a culture of impunity. Even before Trump, CBP’s arrest and misconduct rate was five times higher than other law enforcement agencies (it may have had a higher per capita crime rate than undocumented immigrants), and it was probably the nation’s deadliest federal law enforcement agency. The CBP’s union even called an award for CBP officers who avoided the use of deadly force “despicable.” Trump’s giant new hiring surge-- and, you know, everything else-- will likely lead to even lower standards, more impunity, and a worse culture for the law enforcement agency.
The Democrats’ demands for reining in the behavior of the Department of Homeland Security are almost embarrassingly weak, in many cases amounting to nothing but “the Department of Homeland Security should follow the law” [The Watch]. The DHS needs a warrant to enter people’s houses? Officers should be unmasked and have their names and badge numbers prominently displayed? The DHS is not allowed to racially profile people or use force in unreasonable ways? This is basic fucking shit.
Six percent of Trump voters don’t admit to voting for him in the last election [The Argument].
“It’s impossible to say when IT will HAPPEN. But it can’t be too long until IT HAPPENS. Looking at the data (age, high-stress job, cardiac history), it is statistically plausible that IT will HAPPEN in the next thirty-six months. Eighteen, if you factor in hamburger consumption and all the weird bruising. Of course, it doesn’t feel right to want IT to HAPPEN. And it’s obviously not okay to try to make IT HAPPEN. That’s not what this is ABOUT, just to make things CLEAR LEGALLY as far as VARIOUS AGENCIES are concerned. But regardless, IT is going to HAPPEN. So you’re allowed to think about IT.” [McSweeney’s Internet Tendency]
Meta Effective Altruism
The optimizer’s curse is the mathematical fact that, in any situation where your estimate of the best course of action is noisy, you will almost always overestimate the effect of whatever course of action you think is best-- even if your estimate is unbiased [Effective Altruism Forum]. The best “speculative” intervention (i.e. interventions you’re very uncertain about) will tend to look better than the best “grounded” intervention (i.e. interventions you understand well), because the increased uncertainty offers more opportunity for noise to confuse the estimate. On some reasonable assumptions, choosing a grounded intervention with lower apparent expected value has higher expected value than choosing a speculative intervention with a higher apparent expected value. The optimizer’s curse might provide a justification for the common intuition that you should favor grounded interventions over speculative interventions with a higher expected value.
The Centre for Effective Altruism has been accused of hostile workplace environment sexual harassment [Effective Altruism Forum]. Senior staff allegedly circulated a document containing speculation about an employee’s trauma, mental health, and personal life, including a sexualized description of her rape. As of when I wrote this post, CEA hasn’t responded. I find this allegation very credible: Fran cites concrete facts, such as an outside investigation that found that Fran was sexually harassed, which would be easy to disprove if the allegation were false. If the allegation is true, I’m extremely disturbed by CEA’s sexism, ableism, and poor judgment.
Policymaking
Jon Stewart thinks that economics is an excuse to let rich people do whatever they want [The Argument]. His attitude both betrays his own past beliefs about the importance of nuance and serious engagement with people’s ideas, and reflects a wider skepticism about economics as a field.
The best reason to be pronatalist is that people want more babies than they actually have [Lyman Stone]. In a liberal society, you’re not going to convince everyone to do something they don’t want to do because it’s good for society. You might be able to convince people to implement concrete policies that make it easier for people to do what they want to do. Interesting fact: in Sweden, IVF failure increases depression by 20-30%-- this is a huge fucking effect size.
Related: Why do Koreans have so few children? [Works in Progress] To some extent, Korea shares in the same trends as other developed countries: conflict between motherhood and having a career; intensive investment in a small number of children instead of chiller parenting about many children; the decline of marriage. But Korea supercharges these trends: Korean parents spend a fifth of their disposable income on tutoring, which can run from 7am to 2am in high school. Further, Korea also has the unique problem of having a very successful antinatalist campaign from the 1960s to 1980s, which permanently shifted norms about how many children to have.
The story of the origins of postliberalism is WILD [The Argument]. Apparently Patrick Deneen, the founder of postliberalism, believed some pseudoscience about peak oil which made him reject liberalism because he thought liberalism couldn’t deal with peak oil, and then when it turned out that peak oil isn’t real he kept all the same arguments and swapped out “peak oil” with “vague cultural collapse of some sort”.
Agricultural cultures tend to be zero-sum because there actually is a fixed amount of Stuff to go around, and if one person has more Stuff everyone else has less [The Argument]. But liberalism requires that stuff not be zero-sum and that you (immigrants, racial minorities, etc.) can get richer without taking stuff from other people. Artificial scarcities (such as housing) make people illiberal, so liberals ought to particularly oppose them.
Rationality and Mental Health
I really liked this article about what we do when mental illness causes people to be impossible to live with [Psychiatry at the Margins]. It is compassionate both to the people who find those mentally ill people unlivable and to the mentally ill themselves, and doesn’t come to any facile answers. I think a lot of people refuse to admit that mental illness can make people obnoxious, because it feels like there has to be some cap on how obnoxious people can be without it being their fault. Unfortunately, the real world isn’t nearly that convenient! In my personal life, I’ve had to deal with a fair number of people who are profoundly obnoxious because they’re crazy, and I’ve never found a principled way I really like of handling it.
Sometimes people seek out mental health care not because they “really” have any unusual mental condition but because they can’t cope with the conditions of their lives [Psychiatry at the Margins]. On the one hand, a mental health professional is tempted to say “you’re not mentally ill, fix your life.” On the other hand, medications and therapy can (sometimes) actually help neurotypicals who can’t cope with their lives, and it seems wrong to abandon suffering people because their brains aren’t that unusual.
In traditional hunter-gatherer societies, parents don’t validate their children’s feelings all the time [Motherhood Until Yesterday]. In fact, only modern Westerners do this. Constant feelings validation isn’t very helpful because young children don’t necessarily understand what all those words mean, and also find it rewarding when adults pay attention to them. Instead, hunter-gatherer societies grant children autonomy, tolerate bad behavior because they expect children to grow out of it, and are loving and affectionate while ignoring bad behavior.
How to figure out what kind of life makes you happy through thinking through your sensory, social, moral, personal, and safety desires [Untitled Bookcase].
Some kinds of predatory behavior—domestic violence, sexual violence, fraud, theft from acquaintances, ideological abuse—are difficult for society to handle with our current tools [Solar Light]. The social license to operate is tacit societal permission to engage in predatory behavior (for example, by writing off sexual violence as a “miscommunication”). Trying to figure out how to shift the social license to operate is very difficult—for example, cancel culture was a mostly failed attempt to shift the social license to operate.
Particularly Good: When Prophecy Fails, the book that introduced the concept of cognitive dissonance, turns out to have been fraudulent [Journal of the History of the Behavioral Sciences]. Researchers faked messages from aliens, deliberately manipulated the UFO group to get the results they wanted, and even interfered with the child welfare investigation of a group member. The UFO group proselytized well before the failed prophecy, and broke up shortly after the prophecy failed. In reality, religions that make failed prophecies tend to fall apart, not become stronger through cognitive dissonance.
Related: Oliver Sacks’s early books (up until The Man Who Mistook His Wife For A Hat) were exaggerated to the point of being fraudulent [The New Yorker]. While all of his case studies really existed, Sacks made up details (including two savants who could calculate twenty-digit primes) and put his own words in the mouths of his less articulate patients. His later books, which drew more on people who wrote to him rather than his own patients, seem to have been entirely honest. Between Sacks and When Prophecy Fails, I’m very concerned about how much social science fraud has been done by people who don’t happen to have obsessively documented their fraud.
The histories we write about science reflects the different things we want out of a history of science [Asterisk Magazine]. There’s no single objective “history” of science; we tell different stories for different purposes.
Reality Has A Surprising Amount of Detail
Chinese Doom Scroll: Mind games about what you should do if your boss tells you to get liquor from the car and his car doesn’t have any liquor in it.
Prison gangs are a form of governance in overcrowded prisons [Asterisk]. Particularly interesting: gangs have written rules that members have to read before they can join, including bedtimes and taxes on drug sales.
An introduction to Roblox for adults who don’t know what Roblox is [Infinite Scroll]. “More people play Roblox on a monthly basis than play the Nintendo Switch, PlayStation and Xbox combined.” “Roblox is less like a video game, and more like a mall, a Discord, a game engine and a role playing community all jumbled up together.
This is where youth culture is being created. This is where kids are learning to navigate the digital world, where they form new memes, and where they experiment with identities.”
Trad influencer cheats on his spouse with another trad influencer [The Bulwark]. The only person in this scandal who comes out of it looking good is conservative influencer Allie Beth Stuckey, who correctly pointed out that “tradwife” bullshit is often more of a sexual fetish than a reflection of genuinely socially conservative ideals.
Brains may have two parts [Asterisk]: a learning subsystem (that creates a model of the world, creates plans, and predicts what will happen) and a steering subsystem (that teaches the learning system what goals it should have by providing rewards for food, warmth, social bonding, play, exploration, attractive mates, etc.). Thought Assessors connect complex concepts in the learning system (such as “my boss” or “interesting geek” or “Factorio”) with the simple signals of the steering system (such as “social approval” and “sexual attraction” and “play”). Humor, for example, comes from a Thought Assessor that maps increasingly complex concepts to “I am basically safe but a little bit in danger.”
It turns out you can make steak and sunny-side up eggs in the microwave [Telescopic Turnip].
‘Evolutionary mismatch’ is the way that our society doesn’t match the environments humans evolved in [Evolution and Psychiatry]. For example, human mothers are very isolated, without the assistance that would have been normal in the environment of evolutionary adaptedness. In some cultures, mothers provide as little as 25% of childcare in the child’s first year of life.
Korean and Japanese coffee shops [Chasing Sheep]. “Dabang, however, developed a seedy reputation over time due to the popular variant known as ticket dabang. At such establishments, a customer could order coffee and cigarettes to be delivered, with the added feature that the courier was sometimes a woman who would provide sexual services upon arrival.” What.
Particularly Good: How China’s ideologies repeatedly brought the empire back together after it fractured [Mr. and Mrs. Psmith’s Bookshelf]. “The ancient Chinese were famously pluralistic in matters of religion, accepting and integrating strange gods from foreign lands. Perhaps that’s because the only jealous god in their pantheon lived on earth. It was an exact conceptual inversion of medieval Europe: there could be many gods, but only one emperor.”
Fiction
Black Flowers Blossom [Uncanny]: A tender romance between an eldritch horror and a reincarnating occult detective. I love it when people want to fuck the eldritch horror.
Editors don’t want male novelists [Woman of Letters]: a complex little story about sexual coercion, the banality of evil, and being awful but not quite as bad as you could have been.
Elven Language [Goblin Punch]: what would a species of immortal supergeniuses actually write like?
Jennifer’s Daughter [Nightmare]: A freaky little horror piece about a teenage girl raised by a zombie monster. I liked the concrete life details and the pervasive sense of loneliness and alienation; they brought a sense of realism to a supernatural premise.
Eating Bitterness [The Dark]: What if women had mouths on their necks that consumed the feelings of their families, which makes the families happier while injuring the women themselves? Would that be fucked up or what?


Sadly I don't necessarily find the accusations of sexual harassment all that surprising in EA circles. IMO the attitude that your circle is particularly virtuous in some related (too rational, too holy, too woke and aware) is a risk factor. Indeed, I think one of the worst things we do regarding sexual assault and harassment is to present it (as in Epstein) as something only someone truly awful would do and thereby make it that much harder for anyone to even wonder if they or their friends might be doing it.
Having said that, I'd quibble about framing the problem in terms of our tools not being sufficient to handle predatory behavior. I mean it's only a quibble since most of what the piece says seems right and I certainly agree predatory behavior is a very hard problem. One of the things that makes it so very hard is that many 'solutions' can themselves be abused or even make things worse. For instance trying to bar relationships between TAs and students or boss and subordinate often give the predatory individual greater power (you promised to keep us a secret, I could lose my job). But framing it as our tools are inadequate presupposes we should expect them to be better. Is that right? Or is it a very hard problem and we should be surprised we are doing as well as we are?
Personally, I tend to think that the better frame is: given that sexual harassment and assault seem unfortunately common how can we help recognize and avoid doing or enabling that behavior rather than framing it as what tools do us good people need to stop those bad guys. Not to mention the risks that trying to create those tools will make things worse.
To answer the tweet: because humans are evolved to find love stories particularly compelling. Other things being equal a love story is just going to be more interesting to most people and when you make it a competition you look for every advantage. Not to mention that it's much easier to communicate a love story within the limited range of ice skating.