Astral Codex Ten

Chetty

19 min read

Chetty

Article

Chetty is a recurring person in the Astral Codex Ten archive, appearing 2 times across 2 issues between August 09, 2023 and June 27, 2025. The archive places it in contexts such as “Chetty, Deming, and Friedman find that”; “Chetty showed that how (or at least where) your kids are raised can matter”. It most often appears alongside Erik Hoel, India, OpenAI.

Metadata

  • Category: People
  • Mention count: 2
  • Issue count: 2
  • First seen: August 09, 2023
  • Last seen: June 27, 2025

Appears In

Source Context

Recovered passages from the original issue text. When the raw archive preserved outbound links inside the source passage, they are listed directly under the quote.

Links For August 2023
August 09, 2023 · Original source
13: Fact check: was Elvis Jewish? Snopes says yes, but I’m more convinced by this argument for no. [update: commenter TheGenealogian agrees no] 14: Is GPT-4 getting worse? This isn’t absurd; some people claim OpenAI has simplified the model to cut costs (though OpenAI denies this). Matei Zaharia argues yes, but I’m more convinced by the AI Snake Oil blog’s argument for no (h/t Stuart Ritchie). 15: Vox has a good piece about AI company Anthropic. I would quibble that they’re not the only safety-focused or EA-affiliated org, and we have yet to see how truly safety-focused or altruistic any AI company can be while continuing to be an AI company. But granting that it’s all a matter of degree, I agree the degree seems pretty high for them. And NYT also has an Anthropic article. 16: Eliezer bets $150,000 to $1,000 against UFOs being aliens, and gives the same argument I would - it’s unlikely that any civilization advanced enough to travel through space would still be primitive enough to use macroscopic, biologically-piloted craft that sometimes crash. 17: More nails in the coffin of growth mindset. “When examining the highest-quality evidence (6 studies, N = 13,571), the effect was nonsignificant: d = 0.02, 95% CI = [−0.06, 0.10]. We conclude that apparent effects of growth mindset interventions on academic achievement are likely attributable to inadequate study design, reporting flaws, and bias.” I think the older, very-high-effect-size studies were clearly terrible, but I’d still like to look further into the newer, small-but-significant-effect-size-that-makes-a-difference-across-large-groups studies and how they went wrong. 18: Previous work showed that after adjusting for selection bias, “what college you go to doesn’t matter” for average earnings. I was always skeptical of this - are all those rich people sending their kids to Ivies for no reason? Now Chetty, Deming, and Friedman find that: Attending an Ivy-Plus college instead of the average highly selective public flagship institution increases students’ chances of reaching the top 1% of the earnings distribution by 60%, nearly doubles their chances of attending an elite graduate school, and triples their chances of working at a prestigious firm. Ivy-Plus colleges have much smaller causal effects on average earnings, reconciling our findings with prior work. One of the authors, David Deming, has a Substack here where he explains the study in more depth. Like everyone else, this study also finds that rich people are using “holistic admissions” and the de-emphasis of standardized testing to gain an advantage: H/T Nate Silver, who writes: “Not sure how you can look at this data, ostensibly be interested in either meritocracy or equality, and want to move away from standardized tests. It's the subjective measures that are most slanted in favor of the rich kids.” Cf. Erik Hoel. 19: From @data_depot: “In 2002, 48% of Americans said "the govt is run by a few big interests looking out for themselves." 52% said "it is run for the benefit of all people." In 2020, 84% said the govt is run by a few big interests. Only 16% said it is run for the benefit of all people.” Source seems to be here, which reveals 2002 was a local peak in trust in government; maybe because of post-9/11 unity, but even 2000 was 34%, much better than our current 16%. My first instinct is to attribute this to a rise in vulgar Marxism, in the sense of everyone (even conservatives) now being trained to think in terms of an elite class screwing over everyone else (cf my review of Manufacturing Consent). But there was a previous low of 19% in 1994, which doesn’t seem to correspond to anything especially bad going on in the US, so I don’t know. 20: AskReddit: Medical professionals - have you ever had a patient so lacking in common sense you wondered how they made it so far? Linking this because there’s lots of evidence showing that education (as a proxy for intelligence?) is associated with increased life expectancy, and this thread gives you a visceral appreciation of why that might be. 21: The Fall Of [programming help site] Stack Overflow: Looks like a weak downward trend since 2021 I can’t explain, plus a strong downward trend since 11/2022 which must be from ChatGPT. In case you were wondering how AI was affecting programming! (update: probably false, see here, though see also here for evidence of smaller but real decline) 22: This month in culture war topics: London’s Pride parade featured a convicted kidnapper/torturer/rapist/sadist as a speaker, who advocated that anti-trans people should be “punch[ed] in the f**king face” ; the organizers say they stand by her.
Inline links: yes, this argument for no, agrees no, argues yes, argument for no, Stuart Ritchie, Anthropic, an Anthropic article, the same argument I would, More nails in the coffin of growth mindset, Previous work, Chetty, Deming, and Friedman, has a Substack here where he explains the study in more depth, https://substackcdn.com/image/fetch/$s_!VcFl!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F4f08bfe5-ab31-453a-896a-54ef385da7d2_706x900.jpeg, Nate Silver, Erik Hoel, @data_depot, https://substackcdn.com/image/fetch/$s_!S4g-!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff18fa4d5-9ba9-4b86-a058-46246bfc8a4f_536x611.png, here, my review of, have you ever had a patient so lacking in common sense you wondered how they made it so far?, The Fall Of [programming help site] Stack Overflow, https://substackcdn.com/image/fetch/$s_!E7XK!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff8ba7c05-7dbb-4318-9da2-87b00d738ed7_649x518.png, probably false, see here, here, featured, stand by her
Your Review: Alpha School
June 27, 2025 · Original source
“Powered by AI (not teachers).” If all of this makes your inner Bayesian flinch, you’re in good company. After twenty‑odd years of watching shiny education fixes wobble and crash—KIPP, AltSchool, Summit Learning, One-laptop-per-child, No child left behind, MOOCs, Khan‑for‑Everything—you should be skeptical. Either Alpha is (a) another program for the affluent propped up by selection effects, or (b) a clever way to turn children into joyless speed‑reading calculators. Those were, more or less, the two critical camps that emerged when Alpha’s parent company was approved to launch the tuition‑free Arizona charter school this past January. Unfortunately, the public evidence base on whether this is “real” is thin in both directions. Alpha’s own material is glossy and elliptical; mainstream coverage either repeats Alpha’s talking points, or attacks the premise that kids should even be allowed to learn faster than their peers. Until Raj Chetty installs himself in the hallway with a clipboard counting MAP percentiles it is hard to get real information on what exactly Alpha is doing, whether it is actually working beyond selection effects, and if there is anyway it could scale in a way that all the other education initiatives seemed to fail to do. I first heard about Alpha in May 2024, and in the absence of randomized‑controlled clarity, I did what any moderately obsessive parent with three elementary-aged kids and an itch for data would do: I moved the family across the country to Austin for a year and ran the experiment myself (unfortunately, despite trying my best we never managed to have identical twins, so I stopped short of running a proper control group. My wife was less disappointed than I was). Since last autumn I’ve collected the sort of on‑the‑ground detail that doesn’t surface in press releases, or is available anywhere online: long chats with founders, curriculum leads, “guides” (not teachers), Brazilian Zoom coaches, sceptical parents, ecstatic parents, and the kids who live inside the Alpha dashboard – including my own. I hope this seven-part review can help share what the program actually is and that this review is more open minded than the critics, but is something that would never get past an Alpha public relations gatekeeper: Starting Point: My Assumptions: how my views on elite private schools, tutoring and acceleration shaped the experiment (and this essay). WHAT is the existing education environment.
They keep the kids motivated so they put in the daily effort and don’t get burned out What Alpha is doing is not rocket science. They are just “following the science” for what has been proven to work, and then designing a school around the best way (or “a way”) to deliver that science - personalized instruction, mastery focus, spaced repetition and incentives. It should not be too surprising that when it all comes together it spits out measurable results. But will it hold? Part Six: A Response to Bryan Caplan "When the data and anecdotes disagree, the anecdotes are usually right. It’s usually not that the data is being miscollected. It’s usually that you’re not measuring the right thing." — Jeff Bezos (on multiple occasions) Not only does Bryan Caplan convincingly argue that education is mostly signalling in his book “The Case Against Education”, he goes even further to pour cold buckets of water on aspirational parents in his book “Selfish Reasons to have more Kids”. In that latter book he makes a compelling case with unimpeachable data that how kids turn out is almost entirely due to their genes plus “non-shared environment” (i.e., random things not having to do with parenting). According to Caplan helicopter-parenting does not hurt your kids, it is just a waste of everyone’s time (and maybe their enjoyment during their childhood). You might be able to influence some of your kids' behavior in the short term, but once they become an adult and move out of your house they will revert to the biases of their genes. As Caplan says, the most important parenting decision you can make that will affect how your kids turn out is your choice of spouse (or more accurately your choice of the genes you use to build your kids). Caplan does put one caveat on his data: range restriction. He admits that all of his adoption studies focus on middle class Americans (and Europeans). He is the first to admit that if you take a baby out of extreme poverty in the developing world and raise him in a middle class American family, he will have better economic outcomes than if you leave him in rural Mauritania (see his “Open Borders” book). He may even grant that moving from the poorest broken families in America to the middle class also may make a difference – since all the data available comes from families who were approved by administrators as acceptable to raise adopted kids. But is the same thing true when you move from the middle of the bell curve to the right? When the Data Set Gets Bigger Raj Chetty’s neighbourhood-impact study cracked the range challenge open. Chetty had access to all IRS filing data for generations. He was able to focus on families with multiple children that moved to significantly different zip codes, and follow those children over extended periods of time. By having millions of data points he could tease apart the impact of moving to a “better” zip code for older vs younger siblings. The younger sibling had the same family environment (and 50% of the same genes), but some number of more years in the “better” neighborhood. Chetty found that better neighborhoods made a difference to long term outcomes. But isn’t the neighborhood where a family lives in a “shared environment”? Clearly some adopted families lived in better neighborhoods than others? Why didn’t Caplan’s adoption studies pick that up? I think part of the answer is noise. Chetty had millions of data points vs hundreds of thousands for the adoption studies. But mostly I think the reason Chetty found this impact while the adoption studies did not is that he was looking for different things. No one took the adoption studies and grouped the zip codes as the relevant input variable. As Bezos says, the data wasn’t miscollected, they were just looking at the wrong things. So what does a good zip code look like? Chetty summarizes a good zip code as: Short commutes
Inline links: Bryan Caplan, The Case Against Education, Selfish Reasons to have more Kids, Open Borders, neighbourhood-impact study
Low single mother rate He summarizes that as a place of “economic connectedness” – where adults are connected to each other and to the broader community. A lack of those five elements are not bad per se, but they are correlated with a community where people are not interacting with each other as much as they are in communities where the metrics are reversed. Chetty frames it that kids are influenced by the other adults in the area they live in. But I have another hypothesis. Rather than: Other parents → Your kids Perhaps the causation runs from: Other parents → You → Your kids Maybe it’s not other parents' style of parenting that is influencing your kids (how?) but rather when you spend time around other parents their parenting style rubs off on you and how you parent your kids. Influence like that will not get picked up in Caplan’s adoption studies (which focus almost on how parent characteristics get passed on to genetic vs adopted children’s characteristics), but it is a potential signal that maybe parenting choices do matter. Maybe we were just looking at the wrong data. Pre-registered Genius Experiment We now have two data sets that don’t contradict directly, but do point to opposing conclusions. It would be great if we could test this with a pre-registered randomized control trial. That is not going to happen in our current culture. But enter Laszlo Polgár, who volunteered his own children as the test subjects. (Scott’s 2017 review of Polgar’s book here) Before his children were born Polgár publicly announced he would raise them to be geniuses. He initially considered training them to be genius artists, writers or mathematicians, but decided those fields were not objective enough. It would be too easy for critics to dismiss his future children’s achievements and “not genius” no matter what they accomplished in those fields. So he chose a field that was considered both “driven by intelligence” that had clear, objective measures: chess. Then he called his shot. By 1989 all three girls received their first “GM norms” (a GM norm is finishing a tournament with a elo score of at least 2600; 27 norms are needed to make grandmaster). Two went on to become grandmasters - the 3rd and 4th women to ever achieve that title. One ranked in the top 100 (all genders) at age 12 – she peaked at #8 in the world. The other became the top-rated woman in the world at age 15. Polgar showed that you could take kids, at least kids with “good enough genes”, and turn them into world champions through the right education methods. One might think this would be “case closed”, but even as the Polgar sisters were achieving these feats people were saying that these girls must have been “naturally gifted”. They clearly had bright parents, but does anyone think that if they had been adopted into a random middle class American household they would have still become chess geniuses? Or world class in anything at all? When Polgar was challenged on exactly that, he wanted to repeat the experiment by adopting a “black child” and doing it again. Unfortunately his wife talked him out of it. Even if he had adopted a child and turned him into a genius, that would just be one data point – it would not show up in Caplan’s adoption studies. It would be a case of the anecdote and the data disagreeing. Which do you choose to believe? Aristocratic Tutoring It would be great if we could find more examples of Polgar’s model. While I could not find any other “called shots”, one could go back and look at the childhoods of geniuses to see if there is anything to find. That is what Erik Hoel did in his series of posts on “Why we stopped making Einsteins” (post 1, post 2, post 3; Scott’s response). Hoel argues persuasively that, when biographies of their childhoods exist, the geniuses of the past were almost all given 1:1 tutoring. There must have been many aristocrats in the past that were given 1:1 tutoring who never amounted to world-class genius, and many world-class geniuses who got there without 1:1 tutoring, but it does seem to put the thumb on the scale. Benjamin Bloom would agree. Benjamin Bloom quantified Polgar’s hunch in 1984, just eight years after Polgar’s last daughter was born. He ran a RCT where some students were taught normally and others given 1:1 tutoring. He found that the average tutored child improved by two standard deviations over the control: “The average tutored student was above 98% of the students in the control class” and “about 90% of the tutored students ... attained the level reached by only the highest 20% [of the control]”. He called his finding the “Bloom’s 2 sigma problem” Why would this discovery of the secret sauce that could turn the average student into a genius be a problem? Because Bloom saw no way to scale it. Clearly we can’t give every kid in the world a personal 1:1 tutor. We had the solution that would revolutionize everything, but it was just too expensive. Where does that leave us? Caplan showed that, within the normal range, nothing you do in education or parenting matters. …But Chetty showed that how (or at least where) your kids are raised can matter. …Polgar showed that intense 1:1 tutoring from a young age can create world-class geniuses …And Bloom showed that 1:1 tutoring can work for almost everyone, improving performance, if not to world-class levels, still two standard deviations above the alternative. Caplan is still mostly right—if you hover in the complacent middle of American schooling. But Chetty hints that context nudges outcomes, Polgár proves that deliberate, early, personalised instruction can manufacture prodigies, and Bloom tells us it lifts the average child by two sigmas. Alpha’s claim is that software‑mediated, 5:1 tutoring narrows that two‑sigma gap for a price mere mortals can (barely) contemplate. Whether that vision survives contact with budgets, regulators, and human nature is the question for section seven. Part Seven: Scaling Weird A month into our experiment in Austin we were at a neighbor’s backyard pool party (a fringe benefit of moving to Austin: there were backyard pool parties in early November). I was in conversation with a couple that I had just been introduced to. He asked why we moved to Austin, “Was it for your job?” “No. Actually we moved for a school for the kids.” Their faces expressed a combination of confusion and shock. It wasn’t the first nor the last time. Everyone is confused at why we would move across the country to send our kids to a new school, “They don’t have good schools where you come from? How much does this school cost?” Those two questions frame Alpha’s biggest risks when it comes to scaling. Their biggest challenges going forward are not going to be pedagogical. They are going to be sociological and economic. The Economic Problem Alpha is much cheaper than a Victorian Governess, but it’s not cheap. As mentioned in this review more than a few times, Alpha’s flagship campus charges $40,000 a year— roughly 3-4× what the other top-tier private elementary schools in Austin ask. Yes, that figure is all‑in: every Chromebook, every afternoon workshop, even the spring junket to Poland to beta‑test the platform with Ukrainian refugees is baked into tuition. There are no gala auctions or booster fees waiting in tall grass. Still, $40k is a hard swallow when the local Christian school will take your child for eleven. Worse, the number almost certainly fails to cover costs. Recall that guides start at $60k, rise to $100k on promotion, and the five “head guides” each earn $150k. At the five‑to‑one student‑to‑teacher ratio Alpha runs, those salaries alone suck in half the revenue from a twenty‑kid cohort before you’ve paid the rent, the head of school, the company executives, the curriculum designers, the engineers that are building the 2-hour platform and AlphaRead, the workshop costs (or the trip to Ukraine) or the marketing expenses (MacKenzie has a very well produced podcast, and I see a lot of ads for the school on Facebook now that we live locally). Compared with aristocratic one‑to‑one tutoring, forty grand is a steal. But $40,000 is still Lamborghini kindergarten – and even at those prices it is still burning through Joe Liemandt’s cash pile. Alpha’s answer to eventually solving the economics seems to be two fold: (1) Get enough scale that the fixed costs (like the learning platform) become a rounding error on overall costs, and (2) pull out the “non-essentials” at many of the campuses to get the marginal cost well below $10,000 per student. Whether they will be successful is still in early innings. The homeschool product beta is limping along with 1x learning, and the Arizona Charter doesn’t open until autumn 2025. Whether Alpha retains its magic without $150,000/year guides with 5:1 teacher:student ratios and generous bribe incentives programs, remains to be seen. The Weirdness Problem When Bryan Caplan writes about the signaling theory of education, he lists three signals that schools send to employers: Our students are smart
Inline links: He summarizes that as a place of “economic connectedness”, Laszlo Polgár, Scott’s 2017 review of Polgar’s book here, post 1, post 2, post 3, Scott’s response, Benjamin Bloom