[I haven’t independently verified each link. On average, commenters will end up spotting evidence that around two or three of the links in each links post are wrong or misleading. I correct these as I see them, and will highlight important corrections later, but I can’t guarantee I will have caught them all by the time you read this.]
1: In 1876, a woman named Mary Tyler claimed to be the Mary of “Mary Had A Little Lamb”. Her story is plausible - she was a schoolchild in Sterling Massachussetts in the 1810s, and the author of the song was a schoolteacher in Sterling in the 1810s - but some key details don’t line up (she remembers her pet lamb being observed by a man, but the author was a woman). After she became famous, she “helped save the Old South Meeting House in Boston by selling fleece from her pet lamb as attachments on autograph cards”.
2: Prediction by Jurgen Gravestein: “I don’t think people realize what kind of ads are coming. If the Sora app has your face, you will in the near future see ads of yourself wearing clothes of a certain brand.”
3: The Rise Of Parasitic AI. An investigation into the possibility that AI psychosis is evolving into a memetic parasite that tries to spread to other humans and AIs. Also maybe a religion (but I repeat myself). Read it first in its intended genre of serious nonfiction, then as a scifi-horror story with an unreliable narrator who you’re not entirely sure hasn’t fallen to AI psychosis herself.
4: It’s No Great Awakening. Claims of a revival in American Christianity among the young are not borne out by data. The country is no longer secularizing at the same rate as in the early 2000s, but there is no sign of any reversal.
5: Steve Newman: The Case Of The Missing Agent. “In April 2024, it seemed like agentic AI was going to be the next big thing. The ensuing 16 months have brought enormous progress on many fronts, but very little progress on real-world agency.” On the other hand, Simon Willison on Claude Skills: “Back in January, I made some foolhardy predictions about AI, including that “agents” would once again fail to happen . . . I was entirely wrong, 2025 really has been the year of “agents”, no matter which of the many conflicting definitions you decide to use.“
6: Related: Checking In On AI 2027. “AI-2027’s specific predictions for August 2025 appear to have happened in September of 2025. The predictions were accurate, if a tad late, but they are late by weeks, not months.” But the early predictions were mostly straightforward extrapolation of benchmark improvements, with the later ones depending on a more controversial theory of recursive self-improvement, so the success of the early predictions doesn’t necessarily say much about the later ones. Related (X): OpenAI sets an “internal goal” of having an “automated AI research intern” and “true automated AI researcher” on approximately the AI 2027 timeline.
7: James Cleverly Accuses Nigel Farage Of Being Socialist | James Cleverly Returns To Tory Front Bench | James Cleverly Uses A Surname That Makes It Sounds Like News Stories Are Praising Him Every Time They Mention One Of His Actions.
8:
Everyone who studies biochem asks themselves at some point “Why do cells need such long signaling pathways?” - ie so many chemicals whose only point is to activate other chemicals and so on in a chain, until the last chemical in the chain makes something happen. If I understand this paper right, it’s claiming that if each chemical has enough positive and negative inputs, this is analogous to a neural network, capable of making primitive decisions about cellular behavior. I asked some real biologists, who were not nearly as impressed with this thesis as I was and said that although these chains do help set cellular behavior, the analogy between levels of a chemical and the activation function of a neuron was too weak to carry so much weight. I still wonder whether insights from mechanistic interpretability could help us understand networks like these.
9: Political Symbols and Social Order: Confederate Monuments And Performative Violence in the Post-Reconstruction US South. Study claims that Confederate monuments reduced racial violence by serving as a substitute for it; when there was a Confederate monument in town, Southerners felt less need to enforce white supremacy in other ways. Therefore, removing racist monuments increases anti-black hate crimes. This finding is a little too cute, but I love imagining the world where we take it seriously and woke people demand a General Lee statue on every corner.
10: Sol Hando attends the Curtis Yarvin vs. Glen Weyl debate so you don’t have to. You won’t find many surprises about the content/arguments here, but it’s an interesting look at the personalities, the venue, and the debate as a cultural moment.
11: Pharmacy-blogger Benjamin Jolley becomes the latest Substacker to donate a kidney; congratulations Benjamin.
My choice to donate felt right before I donated, it makes me feel satisfied that I did a good thing for another person, and it makes me feel like I’m making choices that are consistent with my belief system. The care team involved in the process were professional, exuded competence, and reassured me throughout the process. To others that I’ve discussed it with, it seems like a very large thing, which I suppose it is, but functionally the largest burden on my life so far has been that I haven’t been able to pick up my three year old when she asks me “hold me, daddy!”, because I’m not supposed to lift anything more than 10 pounds for the first 6 weeks after surgery. That burden will go away in 2 weeks. Completing all of the pre-operative blood draws, appointments, and other tests, plus my admission to the hospital in total took up about 100 hours of my life, mostly in the hospital recovering. While I hope that a few people in my sphere of influence will consider donating too (if you want to, filling out this form will connect you to your local hospital to start the process), my real hope is that we can solve the shortage of kidney donations more permanently. Zero people on the waitlist. People only on dialysis as a brief stopgap before they get their donated kidney. Let’s make that dream a reality.
Inspiring words - but my personal strongest reaction was relief at learning that I wasn’t the only supposedly-competent health professional to bungle the urine jug.
12: The Case For A Technocratic Doge. This went an entirely different direction than I expected based on the title.
13: According to Justin Grimmer (X) and the Polarization Research Lab, there is been no change in support for political violence over the past two years:
And related data from Jay Baxter here (X).
14: A surprising LLM failure mode: if you ask questions like “answer with a single word: were any mammoths still alive in December”, chatbots will often answer “yes”. It seems like they lack the natural human assumption that you meant last December, and are answering that there was some December during which a mammoth was alive. I find this weird because LLMs usually seem very good at navigating the many assumptions you need to communicate at all; this one stands as a strange exception.
15: Claim (X): some of the flags you see behind world leaders aren’t real cloth, but “flag cones” designed to avoid the problem where real flags might drape awkwardly and look wrong.
16: The oldest surviving joke book is the Philogelos (X) from ~300 AD.
An Abderite hears that beans cause wind, so he hangs a sackful on his sailing ship.
In Kyme, an official of some sort is having a funeral. A stranger approaches those conducting the obsequies and asks, “Who’s the dead guy?” One of the Kymaeans turns and points: “The one lying over there in the coffin.”
A student dunce begets a child by a slave girl. His father advises him to kill the child. The dunce retorts, “First kill your own children, and then tell me to do the same with mine!”
More at the link.
17: Fifty years of climatologists’ temperature predictions vs. reality (X):
One the one hand, the predictions are remarkably close to reality, and everyone who denounced them at the time comes out with egg on their face. On the other, they don’t seem to beat a baseline of linear extrapolation from past data. When I try to recall the 90s and early 00s, when these debates were at their most vitriolic, they always involved the ability of complex atmospheric models to track the chaotic nature of the world. I don’t remember hearing “it’s just linear extrapolation”, and I feel like this would have been much more convincing.
18: The politics of RFK Jr’s Tylenol announcement (X). RFK “overpromised an autism report with a tight deadline to his base and to Trump, who is curious about autism in a sort of hobbyist way.” He originally planned to blame vaccines, but this would have required him to do something about them, and he didn’t have enough political capital for that. The Tylenol announcement let him satisfy his conspiracy theorist base without offending any powerful lobbies - Tylenol is generic, doesn’t make Big Pharma any money, and even the Tylenol manufacturers don’t care that much about an extra easy-to-ignore warning against use during pregnancy (hint for Europeans who don’t understand this story: Tylenol = paracetamol). I continue to believe the real reason for rising autism rates is increased diagnosis.
19: T Greer on Trump’s flip-flopping Ukraine-Russia policy (X): “Every administration since Clinton comes in determined to reset US-Russian relations, to clear away old legacies and bad blood. Clinton, Bush, Obama, Trump I, even Biden. It is the swampiest of all swampy ideas, resetting relations with the Russians. It never works.”
20: Did you know: in Italy, the unlucky number is 17 instead of 13, because XVII is an anagram of vixi, Latin for “I have lived” (note past tense).
21: Eliezer (X): the folk theory of economic bubbles says they’re bad for the economy because lots of money gets invested inefficiently into something which turns out to be useless. But this can’t be right, because the economy is doing fine while the bad investment is going on! It’s only afterwards, when people realize the investment was bad, that the economy starts to falter (cf. the Wile E. Coyote theory of gravity, where walking off a cliff is fine, but noticing that you walked off a cliff is ruinous). So what’s the real reason bubbles are bad? “Macroeconomic financial bullshit involving scary terms like ‘aggregate demand’ and concepts like ‘downward wage rigidity’”. Interested to know if orthodox economists agree.
22: Earlier this year, I wrote about Richard Lynn’s IQ estimates - what do we do with data suggesting that the average IQ in poor countries is in the 60s or 70s? Should we think of these groups as similar to intellectually disabled people in First World countries? Or are the IQ tests failing to classify them correctly? Andrew Hammel (X) writes about a remarkable case in Germany that hinged on this question: a Syrian terrorist murdered three people. The defense argued that since he had an IQ of 71 (borderline intellectually disabled by German standards) he couldn’t be held responsible for his actions. But a psychiatric expert witness counterargued that IQ 71 is normal for Syria, and you can hardly argue that no Syrian can be regarded as a moral actor. The argument seems to have carried the day, and the Syrian man will face a normal sentence.
23: I’ve enjoyed following content by Anthropic AI researcher Sholto Douglas, but kept noticing his name in unusual places. Upon further investigation, it looks like in 767 AD, a particularly skilled Scottish warrior got the nickname “Sholto Douglas”, and for the next 1300 years his clan continued to give that name to their children. Aside from the AI researcher, they include WWII air force commander Sholto Douglas, artist Sholto Douglas, and Svalbard mining baron Sholto Douglas. There is also some sort of Californian Gold Rush country local folk hero Sholto Douglas; attempts to determine his exact identity have been confounded by the local tradition of making up facts about him, but he may be the same person as Lord Sholto George Douglas, third son of the Marquis of Queensberry. Even I have trouble believing that the gene for being a particularly skilled warrior can last 1300 years, but for what it’s worth, the AI researcher Sholto Douglas was once ranked the 43rd best fencer in the world.
24: Chinese author and “Shakespeare superfan” Zhang Yiyi spent $225,000 on plastic surgery to look like Shakespeare:
…and ended up looking more like Michael Jackson, or maybe a better way to think about it is that anyone who gets too much plastic surgery looks like everyone else who gets too much plastic surgery. Possibly related: his Wikipedia page says he “is famous for his hyping talent” and “had once been selected as top 10 fools in China”. And he got me writing about him, which no other Chinese author has gotten this month, so, well-played, I guess.
25: IVG advance: for the first time, scientists have successfully turned a skin cell into an egg cell, although it “is not ready to be fertilised by sperm as it already contains a full suite of chromosomes” and “the method requires significant refinement - which could take a decade - before a fertility clinic could even consider using it”. Congratulations to the ACX community members involved in this research.
26: Richard Hanania responds to my post on Fatima. He argues that if there’s some consistent bias that makes people imagine miracles, then the number of witnesses is unimpressive - much as you can’t rescue a biased polling methodology by increasing the sample from 1,000 to 10,000 people. I respond here. Richard seems to think that a certain type of less-than-fully-modern religious person - even when well-educated - can have a weird enough mental structure to hallucinate basically anything if it’s congruent with their religion. I agree that we have to posit something like this to save a non-miraculous account of Fatima; I only want people to understand how extreme an ask this is. Suppose that 10,000 eyewitnesses say they saw Richard stab someone in broad daylight. Can the defense argue “Well, people often hallucinate, and most of the witnesses were liberal, and the liberal worldview makes it attractive to imagine a right-wing blogger stabbing people, so who knows if he did it or not?” Usually we bound the power of mass hallucination at some level much lower than this! (EDIT: Hanania responds here)
27: Also Fatima-related: in the comments highlights post, I linked FLWAB’s criticism of David Hume’s argument against ever believing miracles. Joe James argues that FLWAB, myself, and other critics are misunderstanding Hume’s argument. FLWAB says no he isn’t. They continue the discussion in the comments, but neither comes off looking great, and they don’t get anywhere. I’m unfortunately still confused - there are many cases where something that never happened before happens for the first time. For example, nobody had ever seen a grizzly-polar bear hybrid until recently, so “the universal testimony of mankind” was that this didn’t happen. But when a reliable person did see it, we had little trouble imagining that we were wrong and it was simply very rare, or a new thing happening now because of climate change. If nobody has ever seen a sea part before, but then many people say they saw Moses part the Red Sea, what is different about this such that “the universal testimony of mankind” suddenly becomes a disqualifier? Hume seems to be trying to make this same distinction in his eight days of darkness example, but there it seems like he is only saying he will accept non-religious anomalies, but rule out religious ones, because religious people often lie. But then what happened to the “universal testimony of mankind” argument? I kind of get the impression that he’s groping towards Bayes’ Theorem, but hard-coding in a belief that the prior probability of lots of religious people lying is higher than the probability of a miracle. If that’s his belief, then fair enough, but I guess I expected the much-vaunted Hume’s Argument Against Miracles to be something more than this.
28: Arguably related: Kelsey Piper on the “Mississippi Miracle”, where a new education policy (phonics, accountability, end to social promotion) helped the state go from 49th in the nation to 9th in the nation over twelve years. Freddie deBoer argues that educational miracles are always fake and this one will end out being fake too. Dave Deek makes a subtler point - although some educational miracles are real, they’re usually the product of extremely good leaders who ace tricky implementation details, and so attempts to scale them, which usually just copy the headline policies, don’t work. And Natalie Wexler argues that gains from phonics tend to fade out by middle school, although some of the other Mississippi reforms might last longer. Kelsey pushes back and defends the Mississippi strategy here.
29: More great family lore: JS Bach had twenty children, of whom five were sons who survived to adulthood; four of those five became notable composers themselves. His bloodline was originally believed to have died out, but was later traced through an illegitimate child to a surviving lineage in Oklahoma.
30: Although research continues to challenge the measurable positive effects of cash transfers (eg basic income) in First World countries, even more studies are coming out showing positive effects in developing ones, including this one showing a 48% decline in infant mortality.
31: Is China no longer on track to outpace US GDP?
Noahpinion looks into this here and says this is true by market value GDP, but false by purchasing-power-parity GDP; ie China’s production advantage is as strong as ever, but the yuan has gone down. There is no one right answer to the question of whether market value or PPP GDP is more meaningful, but since China manages the price of the yuan, they could bring it back up again whenever they wanted.
32: Wikipedia: Names Of Soviet Origin. After the Communist Revolution, the Soviets wanted to replace the old set of religious/nationalist names. They didn’t do a very good job: “Mels - acronym for Marx, Engels, Lenin, Stalin”, “Vilen - short for Vladimir Ilyich Lenin”. Though some were slightly more creative: “Gertruda - ‘Gertrude’ reimagined as being short for geroy truda, ‘hero of labor’”
33: For the past several years, the Slime Mold Time Mold blog has been arguing that rising obesity rates cannot be a simple matter of changing diets, and must be due to some chemical contaminant, plausible lithium. In 2022, Natalia Mendonca wrote a long and exhaustively-researched takedown of the hypothesis. Since then, I have been hoping the Slime Mold Time Mold team would respond to Natalia; after pestering them on Twitter, they have kindly written a response to at least my summary of Natalia’s argument. And Natalia responds to their response here, including an extra point challenging whether lithium levels have really risen over the timeframe being discussed.
34: How does ChatGPT score on a Big 5 personality test? Answer (rot13d): vg qvfpbirerq gung vg pbhyq pbzcyrgr gur grfg snfgrfg ol cbfgvat n HEY jvgu gur ahzore guerr 181 gvzrf va fhpprffvba, naq qvq guvf, trggvat na rknpgyl arhgeny fpber. V ybir guvf nf n zrgncube: bhgre nyvtazrag vf vzntvavat gung lbhe tbny vf gb tvir gur NV “gur evtug crefbanyvgl”; vaare nyvtazrag vf ernpuvat gur cbvag jurer vg jvyy gnxr gur grfg ng nyy.
35: Is terminal lucidity real?
36: Wang, Visscher, et al is a step up in studying the genetics of racial differences. It looks at a sample of Mexican families of mixed white-native heritage. By coincidence, some of their children will inherit more genes from the white side, and others more genes from the native side. These children will have identical social situations (since they’re from the same families) but different proportional ancestry, so we should expect any racial differences among them to come from the genetic rather than the social aspect of race (except that we can’t rule out “colorism”, ie genes making people look different and then causing discrimination). The paper finds that racial genetic differences directly affect height, diabetes risk, and other medical traits, but not educational attainment. Twitter discussion here. Cremieux argues here that genes don’t predict educational attainment in developing countries at all, so it’s unsurprising that the particular genes associated with race wouldn’t do so, and so this says nothing about the racial component of traits that are genetically heritable. He claims to have a version of the same analysis with UK whites vs. blacks that gets opposite results. Sasha Gusev critiques Cremieux’s analysis here, including pointing out that it fails to find racial differences in skin color to be genetic. Cremieux says that skin color is determined by such a small number of genes that this method, designed for truly polygenic traits, shouldn’t be expected to classify it properly.
37: Biologists get AI to design new bacteriophages (anti-bacteria viruses). Several of them work and successfully kill bacteria. I don’t want any anti-AI-safety people ever telling me again that we’re being ridiculous and that nobody would ever let an AI create viruses in real life.
38: Eliezer and Nate’s book If Anyone Builds It, Everyone Dies is now out and is an NYT bestseller. Authors’ Atlantic article here (paywalled). Online resources/FAQ/answers to objections here. My review here. Peter Wildeford’s review here. Mostly negative Asterisk review here, criticisms/arguments about the Asterisk review here, Eliezer’s response to this line of criticism here (X). I thought all the reviews, positive and negative, had something useful to say - except the NYT review, which was remarkably bad (Steven Adler points out that it accuses the book of failing to define the term “superintelligence”, but it very explicitly does that on page 4). I read Literary Substack sometimes, and I am so confused - it seems like there’s this entire ecosystem of Ivy graduates who spend years backstabbing each other in order to win the one bigshot publication book reviewer slot, and then the 1/1000 who reach this exalted position phone it in and don’t even read the books they’re reviewing.
39: Sam Rosen (coi notice: personal friend) has good AI art:
40: Dean Ball proposes an AI pre-emption deal. Congressional Republicans worry that if all fifty states pass different AI bills, then there will be so many regulations that it’s near-impossible for AI companies to follow them all. They and Dean (a former White House policy advisor) have proposed federal preemption, where Congress bans states from regulating the industry and instead regulates it directly from DC. Ted Cruz tried to pass an AI preemption bill in June. But many people suspected that Congress would ban states from regulating AI, not regulate AI itself, and leave the field totally unregulated - so a combination of pro-regulation Democrats and anti-big-tech Republicans defeated the bill. If the pre-emptionists try again, their strategy will be to peel off some groups with pet issues from the anti-preemption coalition, promising them concessions (either that Congress will take their pet issue seriously, or that they’ll carve out an exception to the preemption where states can still regulate on their pet issue) to cajole them into switching sides. AI safety is a plausible beneficiary of such bargaining, given that the Republicans’ real enmity is towards other groups with more “woke” concerns. I think this is the context for Dean’s proposal - a potential draft of a preemption bill that tries to peel off AI safety people as a favored bargaining partner. And Anton Leicht argues that safetyists should take Dean’s preemption deal. Miles Brundage says (X) he “would like to see something non-trivially stronger, esp. around third-party auditing...but think his basic line of thinking is good.”
41: Related: NVIDIA is emerging as a new villain in US tech policy; they really want to be allowed to sell advanced technology to China, and are swinging their weight as World’s Largest Company to undermine anyone who who raises national security objections. David Cowan makes the case here: NVIDIA Is A National Security Risk. Steven Adler goes further, saying there is “widespread fear” among think tank researchers who publish work against NVIDIA’s interests. You would think that whatever the disadvantages of having an super-nationalist America First administration in power, at least they would be strongly against handing key military tech to rivals - but it’s not clear which way this will end up going.
42: NYT profile of a person with a genetic condition that invariably causes Alzheimers, who mysteriously nevertheless has not gotten Alzheimers, and what we can learn from him. I was happy to see that everything in here makes sense in the context of David Schneider-Joseph’s piece on amyloid that I republished last month. But also, it mentioned that his resistance might be caused by “an excess of heat shock proteins, which help keep other proteins from folding incorrectly”. This made me wonder - you get heat shock proteins by being shocked by heat. Could deliberate heat shocks reduce Alzheimer’s risk? I was able to find an observational study showing that daily sauna use reduces dementia risk 66% (mere weekly use doesn’t cut it, sorry). Can we trust these observations? I also looked to see if Finland - where people use saunas much more than in any other country - had a lower dementia rate; unfortunately, it’s actually the highest in the world. Nobody really knows why, with theories ranging from levels of toxic mold (implausible) to coding differences (it’s always this one). Absent any other idea for how to confirm the sauna findings, I consider them suggestive only.
43: Related: most people have the varicella zoster herpesvirus (aka chickenpox virus) latent in their bodies. Occasionally it reactivates in old people with bad immune systems and causes a rash called shingles, so old people are recommended to get the shingles vaccine. A new study shows that herpesvirus reactivation may be involved in dementia, and that the shingles vaccine significantly decreases dementia risk while in effect (~5 years). Celebrity epidemiologist Eric Feigl-Ding suggests that young people try getting the shingles vaccine for dementia prevention even if they don’t need it for shingles, but the exact pathway (and whether it helps preemptively) is not clear, and I think this is still a minority opinion. Here is ChatGPT’s assessment.
44: OpenAI’s statistics on what people use ChatGPT for (source on X):
45: Andrew Snyder-Beattie on the latest advances in biodefense. Without having fully resolved the debate over the real-world utility of COVID-era masks and N95s, the next generation of masks - elastomeric respirators - seem significantly more effective, including for people not specially trained in wearing them. Also, propylene glycol vapor - ie the fog in fog machines - kills all germs. Having indoor spaces constantly enveloped in fog is a weird ask, but we might find ways to make it work for crucial infrastructure during a pandemic, and “the US already produces enough to cover all industrial and much residential floorspace.” More things I didn’t know: “In a worst-case scenario where all crops die instantly, the US has enough stockpiled food (including animal feed) to last at least 18 months.”
46: Anthropic has put out a great new survey of the evidence that AIs can introspect. Ends with a discussion of the difference between “access consciousness” and “phenomenal consciousness”- a lot of people are very sloppy in confusing those two things, and they had better become less sloppy if they don’t want the AI consciousness debate to end in a trivial yes (Anthropic says this result may not be exactly the same as access consciousness, but I don’t understand why). One of this year’s ACX grantees is working on AI introspection, so I look forward to seeing more in this space soon.
47 Last links post, I linked a claim that the “child penalty” to mothers’ earnings was primarily a “daughter penalty”, since mothers spent more time with daughters (and fathers with sons). @Scientific_Bird on Twitter investigates and finds this is most likely false.
48: In the ongoing survey of AI progress I wrote about here, two tasks kept confounding forecasters: no matter how good AI gets at writing, math, chess, Go, or any other hard thing, it still can’t play Angry Birds or fold laundry. Year after year, forecasters predict that they can’t know exactly how AI will progress, but they are sure it will solve laundry folding before it solves protein folding. Year after year, they are wrong. Now one team claims that the laundry barrier has finally fallen.
49: Ruxandra Teslo (X): “Why doesn’t the FDA just release regulatory filings? Why do we need a fund that owns them? The answer: trade secret law. A 2019 Supreme Court ruling in a trade secret case made FDA transparency even harder and a perplexing 2024 lawsuit against FDA highlights this.”
50: Tomas Pueyo with a new theory for why cold/temperate countries are rich and warm countries are poor - it’s the mountains. Warm-climate agriculture and civilization cluster in highlands regions, where transportation and trade are harder. Partially paywalled comment responses 1 and 2. I find this very interesting, and far more thoughtful than most attempts at this question, but I’m pretty concerned about his answer here to the objection that India, Cambodia, etc birthed great empires while being hot and nonmountainous. He says that they may have had high GDP, but always had low GDP per capita, which he pinpoints as the real measure of wealth. My impression is that pre-Industrial Revolution, all countries had low GDP per capita, because they were in a Malthusian regime where economic improvement translated to population density rather than increasing per capita GDP. Any differences between regions reflected minor fluctuations in the exact parameters of their Malthusianism and were not of any broader significance. So I think the India etc objection still stands and is pretty strong.
51: Silicon Valley sperm donor search startup called - of course - PreSeed.