Astral Codex Ten

Bloom

27 min read

Bloom

Article

Bloom is a recurring person in the Astral Codex Ten archive, appearing 2 times across 2 issues between April 09, 2024 and June 27, 2025. The archive places it in contexts such as “Bloom (2023)”; “See Bloom (2023)”; “See also Bloom (2021)“. It most often appears alongside Brazil, Poland, Scott.

Metadata

  • Category: People
  • Mention count: 2
  • Issue count: 2
  • First seen: April 09, 2024
  • 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.

Highlights From The Comments On The Lab Leak Debate
April 09, 2024 · Original source
This alone isn’t fatal to lab leak. It’s perfectly possible for the lab to leak (let’s say) November 5th, the virus spreads a bit, and then a month later someone goes to the wet market, coughs on a vendor, and starts the officially recognized pandemic. But if that were true, you’d expect (let’s say) 30 cases by early December. Let’s say the wet market vendor was exactly Case # 30. She infected the other wet market vendors, starting a pandemic with an obvious center at the wet market and lots of infected wet market vendors and patrons. What about Case # 29? If they were (let’s say) a barista, how come they didn’t infect people at their coffee shop? How come there wasn’t a second obvious cluster radiating out from a coffee shop, lots of coffee-shop-linked cases, etc? How come there weren’t 30 equally-sized clusters? In order to avoid this, you either need to claim that the wet market was a perfect superspreader location, or that the pattern with lots of cases in the wet market and few-to-none anywhere else was a result of ascertainment bias. Saar made both those arguments during the debate, but I thought Peter rebutted them effectively. 1.4: COVID in Brazilian wastewater Nicholas Halden (blog) writes: What should we make of this study, which found the presence of covid in Brazilian wastewater in late 2019? Consider the doubling times. The study says that scientists working in late 2020 found COVID in samples of Brazilian wastewater from November 27, 2019. This was long before the first detected case of transmission in Brazil on March 13, 2020. Between November 27, 2019 and March 13, 2020 is about 16 weeks, so 32 COVID doubling times. 32 doubling times with no lockdown is enough time for COVID to infect every single person in Brazil. If COVID had infected everyone in Brazil before the first recognized case, we would have noticed. (again, COVID doubling time isn’t exactly invariably 3.5 days, but here we’re talking about numbers big enough that the exact details don’t matter very much) So if COVID was in Brazil on November 27, it must have fizzled out instead of going pandemic. How likely is that? If one person had COVID, it’s not too unlikely - not all COVID cases transmit it forward. If (let’s say) twenty people had COVID, it’s very unlikely - at that point, the law of large numbers takes over; in a freak coincidence, every single patient would have to fail to infect anyone else. So almost certainly fewer than 20 people in Brazil had COVID in November 27. So which is more likely - that somehow 20 people had COVID long before the virus was officially detected, and on a totally different continent, yet somehow a scientist looking through wastewater found the water from exactly those people and managed to detect the virus? Or that there was a sampling error, which happens all the time in these kinds of things? Peter wrote a blog post on some of these issues. He found that there were positive tests from wastewater samples as early as March 2019, which doesn’t fit anyone’s timeline, including lab leakers’. And most of these positives (including the Brazilian sample) contained later strains of the virus with mutations it picked up late in 2020. So these were almost certainly false positives from contamination. 1.5: Biorealism’s 16 arguments Biorealism has a list of sixteen arguments, which he liked so much that he posted it three times in the ACX comments, twice on Less Wrong, twice on Manifold, and about a dozen times on Twitter under multiple account names. Some posts were slightly different from others, but a typical version is: Importantly, Miller incorrectly claimed the N501Y mutation would result from passage in hACE2 mice (mixed them up with BALB/c mice). The major papers Miller relied on have been seriously challenged since the debate. See Stoyan and Chiu (2024), Weissman (2024), Bloom (2023) and Lv et al (2024). Overall the circumstantial evidence makes lab v plausible: Peter admitted getting this wrong during the debate. I think this very minor point about mice mutations was approximately his only mistake in 15 hours of debating, and he admitted it as soon as he noticed. Biorealism somehow heard about this (obviously not through watching the debate, as we’ll see in a moment), then left about 20-30 comments starting with it, under various accounts, on various platforms, as if it somehow discredited Peter. This is making me somewhat less charitable to him and his 16 arguments than I would be otherwise. 1. Chinese researchers Botao & Lei Xiao observed lab origin was likely given the nearest known relatives to SARS-CoV-2 were far from Wuhan. Wuhan Institute of Virology (WIV) sampled SARS-related bat coronaviruses where the nearest relatives are found in Yunnan, Laos and Vietnam ~1500km away. They refuse to share their records. The ancestral viruses of SARS were found equally far from where SARS spilled over into humans, so we know it’s possible (and likely) for viruses to travel that far. 2. Patrick Berche, DG at Institut Pasteur in Lille 2014-18, notes you would expect secondary outbreaks if it arose via the live animal trade. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10234839/ There are constant outbreaks of weird coronaviruses in animal handlers. See eg this paper, which estimates about 60,000 of these per year. None of these ever go anywhere, because the farmers are in rural areas that aren’t dense enough to sustain a high R0, and the epidemic fizzles out after a single digit number of cases. Any early outbreaks of COVID would have vanished into this long and mostly unnoticed list. 3. Molecular data: Only sarbecovirus with a furin cleavage site. Well adapted to human ACE2 cells. Low genetic diversity indicating a lack of prior circulation (Berche 2023). Restriction site SARS-CoV-2 BsaI/BsmBI restriction map falls neatly within the ideal range for a reverse genetics system and used previously at WIV and UNC. Ngram analysis of the codon usage per Professor Louis Nemzer https://twitter.com/BiophysicsFL/status/1667232580255490053?t=IJgitS5cw364ioclzVWxaA&s=19 The SARS2 backbone is very low in CG and CpG. While the 12-nt insert that gives it the FCS is extremely high in both. Almost as if it was some kind of chimera of a consensus sequence and a codon-optimized polybasic cleavage site? https://twitter.com/BiophysicsFL/status/1752800486837678377?t=EpIRgyybJVaPgeMP5xdstA&s=19 https://www.biorxiv.org/content/10.1101/2022.10.18.512756v1 https://link.springer.com/article/10.1007/s10311-021-01211-0?fbclid=IwAR1HMUMtLIAFOFppVasQDeoIAYrVhP8j4YoPO4wnaTOUiKLsllZl_oKryOw Most of this was discussed extensively in the second session of the debate, which I recommend. The CGG-CGG arginine codon usage is particularly unusual but used in synthetic biology. I asked a synthetic biologist about this. He said: » “Nope. I would literally never do this if I was designing a small insert (maybe I wouldn't notice if it happened by chance with ~1 in 25 odds in a naive codon optimization algorithm as part of a larger sequence). High GC% is bad. Tandem repeat is worse. Several other perfectly fine arginine codons. And I wouldn't engineer a viral genome using human codon usage. An engineer would not do it.” 4. DEFUSE full proposal: virus 20% different from SARS1, consensus seq assembled with 6 segments, without disrupting coding seq, BsmBI order, FCS. SARS2: 20% different than SARS1, 6 evenly spaced fragments w BsmBI and BsaI restriction sites, FCS. Jesse Bloom, Jack Nunberg, Robert Townley, Alexandre Hassanin have observed this workflow could have lead to SARS-CoV-2. Work often begins before funding sought or goes ahead anyway. Re: 4 - Also scattered across second section of debate, also not going to retread 5. Market cases were all lineage B. Lv et al (2024) indicates there was a single point of emergence and A came before B. So market cases not the primary cases. See also Bloom (2021), Kumar et al (2022). Peter Ben Embarek said there were likely already thousands of cases in Wuhan in December 2019.https://t.co/50kFV9zSb6 https://www.ncbi.nlm.nih.gov/pmc/articles/pmid/34398234/ https://academic.oup.com/bioinformatics/article/38/10/2719/6553661 There was a Lineage A sample in the market, lab leak proponents just try to ignore/dismiss/conspiracize it away. The first two known Lineage A cases were very close to the market. Lv (is this even a real name? It sounds like Roman numeral? But I guess that’s what you expect in a country ruled by someone named Xi) found some weird COVID variants in Shanghai that might or might not mean anything; you can see some discussion of the implications here, but I don’t think they’re strong evidence either way. If A was first, it means some really weird stuff coincidences have to happen to give us the spread rates and genetic clock data we get, but they’re not necessarily weirder in the zoonosis hypothesis than the lab leak one. The claim that there were “thousands of cases in Wuhan in December 2019” is very easy to disprove by doubling rate arguments like the one above, by the blood bank study mentioned above, by the WHO’s failed case search, and by many other lines of argument. 6. Evidence for lineage A in the market is based on a low quality sample according to Liu et. al. (2023). I really think lab leakers need to decide whether they think China is a sinister actor trying to cover up the truth, or whether they should trust every offhand comment by Chinese government officials as gospel. Dr. Liu doesn’t explain in what sense he thinks the Lineage A sample is “low-quality”, and the Western scientists who I asked about this said they didn’t understand this complaint and that the sample was fine. A Western team re-analyzing the same sample describes it as “conclusively contain[ing] Lineage A.” I think most lab leakers have switched from trying to deny the genetics to claiming that this was “contamination”, which also doesn’t make sense (the sample is genetically very early). Note that aside from this sample, the first two Lineage A cases discovered were both very close to the wet market. 7. Bloom (2023) shows market samples do not support market origin. There is also no evidence of transmission in the claimed susceptible animals elsewhere. https://academic.oup.com/ve/advance-article/doi/10.1093/ve/vead089/7504441 Discussed extensively in my article as well as the first section of the debate. 8. Lineage A and B only two mutations apart. François Ballox, Bloom and Virginie Courtier-Orgogozo note this is unlikely to reflect two separate animal spillovers as opposed to incomplete case ascertainment of human to human transmission (Bloom 2021). Discussed extensively in my article as well as the first section of the debate. 9. Sampling bias. George Gao, Chinese CDC head at the time, acknowledged to the BBC stating they may have focused too much on and around the market and missed cases on the other side of the city. David Bahry outlines the documented bias. Michael Weissman has shown this mathematically. https://journals.asm.org/doi/10.1128/mbio.00313-23 https://academic.oup.com/jrsssa/advance-article-abstract/doi/10.1093/jrsssa/qnae021/7632556 Re: Dr. Gao, see above comment about Chinese officials. See the section Ascertainment Bias below for why I disagree with this specific claim, which also addresses the Michael Weissman argument. 10. Spatial statistics experts show the Worobey claim the market was the early epicentre was flawed. https://academic.oup.com/jrsssa/advance-article-abstract/doi/10.1093/jrsssa/qnad139/7557954 Re: 10 - See Confirmation Of The Centrality Of The Huanan Market Among Early COVID-19 Cases, a response to the paper you cite: The centrality of Wuhan's Huanan market in maps of December 2019 COVID-19 case residential locations, established by Worobey et al. (2022a), has recently been challenged by Stoyan and Chiu (2024, SC2024). SC2024 proposed a statistical test based on the premise that the measure of central tendency (hereafter, "centre") of a sample of case locations must coincide with the exact point from which local transmission began. Here we show that this premise is erroneous. SC2024 put forward two alternative centres (centroid and mode) to the centre-point which was used by Worobey et al. for some analyses, and proposed a bootstrapping method, based on their premise, to test whether a particular location is consistent with it being the point source of transmission. We show that SC2024's concerns about the use of centre-points are inconsequential, and that use of centroids for these data is inadvisable. The mode is an appropriate, even optimal, choice as centre; however, contrary to SC2024's results, we demonstrate that with proper implementation of their methods, the mode falls at the entrance of a parking lot at the market itself, and the 95% confidence region around the mode includes the market. Thus, the market cannot be rejected as central even by SC2024's overly stringent statistical test. I think this response is pretty strong. In one analysis, they show that even though the other paper’s methodology is worse than theirs, if you apply it correctly (instead of inappropriately excluding various cases like the paper’s authors did), the center of all early cases in Hubei province lands on the wet market parking lot. In another analysis, they show that the other paper’s recommended tests wouldn’t have correctly pointed to the offending water pump in the famous John Snow cholera outbreak, but theirs would have. Still, I think it’s useful to supplement fancy statistics with normal common sense, so I recommend just looking at the map of early cases: …and deciding whether you think the assumptions behind a specific statistical test are likely to debunk the idea that cases are centered around the wet market. 11. Wuhan used as a control for a 2015 serological study on SARS-related bat coronaviruses due to its urban location. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6178078/ I don’t know why this point is supposed to matter. If you mean that Wuhan isn’t directly exposed to bats, nobody ever said it was. The zoonotic theory is that wildlife carted in from other areas of China started the pandemic in the wet market. 12. Superspreader events also seen at wet markets in Beijing and Singapore (Xinfadi and Jurong). This was discussed very extensively in the debates, both in section 1 and section 3. Wet markets weren’t “superspreader locations” - in fact, the disease spread no more quickly there than anywhere else. They were the first place in those cities that the pandemic started, due to contaminated animal products. If anything, this supports zoonosis. See also my discussion with Saar on this point below. 13. WIV refuse to share their records with NIH who terminated subaward in 2022. Wider suspension over biosafety concerns. https://www.bloomberg.com/news/articles/2023-07-18/us-suspends-wuhan-institute-funds-over-covid-stonewalling Although WIV has not been especially forthcoming, some of their databases were leaked in various ways and showed that they did not have any viruses capable of transforming into COVID. 14. PLA involvement at WIV and MERS research prior to SARS-COV-2. MERS features several similarities with SARS-CoV-2. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7022351/ I can’t even tell what conspiracy theory you’re trying to propose with this one; if you spell it out I can try to explain why it might be false. 15. SARS1 leaked several times and SARS-COV-2 has leaked from a BSL-3 lab in Taiwan. Agreed that SARS leaked several times. It also spilled over from animals several times. During the debate, a lab leak rate of once per lab per 500 years was proposed (everyone agreed to steelman this by 10x for WIV numbers); I would be interested to know whether anything about the study of SARS challenges that number. 16. Unpublished infectious clone identified from Wuhan contradicting arguments such reverse genetics systems would be published. https://www.biorxiv.org/content/10.1101/2023.02.12.528210v1.full I asked some scientists about this paper and here’s what they told me. Wuhan University sequenced some rice. In the middle of the sequence, there’s an unexpected sequence from a common coronavirus, HKU4. The most likely explanation is that someone else in Wuhan was working on the coronavirus and there was cross-contamination. Plausibly this is Wuhan Institute of Virology, who is known to work with coronaviruses. This is cool detective work, but it’s not clear what it’s supposed to prove. I think some lab leakers are using it to prove that WIV can do reverse genetics, but they admitted this already in a published paper so that’s not too helpful. I think others are using it to prove WIV had “secret viruses” in their catalogue, but the rice virus wasn’t secret, it was HKU4, which is common and which WIV has already published papers about. 1.6: DrJayChou’s 7 Arguments Once again, I cannot stress enough how much better a take you might have on this debate if you watch it. “The first known case predates the market outbreak by a month” - this is not the consensus position. I cannot say for sure what Dr. Chou means by this, but I suspect he’s referring to one of the many claims to this effect that Peter effectively debunked during the debate (Connor Reed, Mr. Chen, the 92 cases, Brazil, etc).
Inline links: blog, writes, this study, wrote a blog post on some of these issues, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10234839/, this paper, https://twitter.com/BiophysicsFL/status/1667232580255490053?t=IJgitS5cw364ioclzVWxaA&s=19, https://twitter.com/BiophysicsFL/status/1752800486837678377?t=EpIRgyybJVaPgeMP5xdstA&s=19, https://www.biorxiv.org/content/10.1101/2022.10.18.512756v1, https://link.springer.com/article/10.1007/s10311-021-01211-0?fbclid=IwAR1HMUMtLIAFOFppVasQDeoIAYrVhP8j4YoPO4wnaTOUiKLsllZl_oKryOw, https://t.co/50kFV9zSb6, https://www.ncbi.nlm.nih.gov/pmc/articles/pmid/34398234/, https://academic.oup.com/bioinformatics/article/38/10/2719/6553661, here, describes it as, https://academic.oup.com/ve/advance-article/doi/10.1093/ve/vead089/7504441, https://journals.asm.org/doi/10.1128/mbio.00313-23, https://academic.oup.com/jrsssa/advance-article-abstract/doi/10.1093/jrsssa/qnae021/7632556, https://academic.oup.com/jrsssa/advance-article-abstract/doi/10.1093/jrsssa/qnad139/7557954, Confirmation Of The Centrality Of The Huanan Market Among Early COVID-19 Cases, https://substackcdn.com/image/fetch/$s_!BNAm!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fffd4cddb-6e3e-41f5-8ef6-ec0b27bec600_626x426.webp, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6178078/, https://www.bloomberg.com/news/articles/2023-07-18/us-suspends-wuhan-institute-funds-over-covid-stonewalling, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7022351/, https://www.biorxiv.org/content/10.1101/2023.02.12.528210v1.full, a published paper, has already published papers about, https://substackcdn.com/image/fetch/$s_!yA9U!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F467dd304-190a-4437-8920-d498c433dffb_1600x960.jpeg
…and deciding whether you think the assumptions behind a specific statistical test are likely to debunk the idea that cases are centered around the wet market. 11. Wuhan used as a control for a 2015 serological study on SARS-related bat coronaviruses due to its urban location. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6178078/ I don’t know why this point is supposed to matter. If you mean that Wuhan isn’t directly exposed to bats, nobody ever said it was. The zoonotic theory is that wildlife carted in from other areas of China started the pandemic in the wet market. 12. Superspreader events also seen at wet markets in Beijing and Singapore (Xinfadi and Jurong). This was discussed very extensively in the debates, both in section 1 and section 3. Wet markets weren’t “superspreader locations” - in fact, the disease spread no more quickly there than anywhere else. They were the first place in those cities that the pandemic started, due to contaminated animal products. If anything, this supports zoonosis. See also my discussion with Saar on this point below. 13. WIV refuse to share their records with NIH who terminated subaward in 2022. Wider suspension over biosafety concerns. https://www.bloomberg.com/news/articles/2023-07-18/us-suspends-wuhan-institute-funds-over-covid-stonewalling Although WIV has not been especially forthcoming, some of their databases were leaked in various ways and showed that they did not have any viruses capable of transforming into COVID. 14. PLA involvement at WIV and MERS research prior to SARS-COV-2. MERS features several similarities with SARS-CoV-2. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7022351/ I can’t even tell what conspiracy theory you’re trying to propose with this one; if you spell it out I can try to explain why it might be false. 15. SARS1 leaked several times and SARS-COV-2 has leaked from a BSL-3 lab in Taiwan. Agreed that SARS leaked several times. It also spilled over from animals several times. During the debate, a lab leak rate of once per lab per 500 years was proposed (everyone agreed to steelman this by 10x for WIV numbers); I would be interested to know whether anything about the study of SARS challenges that number. 16. Unpublished infectious clone identified from Wuhan contradicting arguments such reverse genetics systems would be published. https://www.biorxiv.org/content/10.1101/2023.02.12.528210v1.full I asked some scientists about this paper and here’s what they told me. Wuhan University sequenced some rice. In the middle of the sequence, there’s an unexpected sequence from a common coronavirus, HKU4. The most likely explanation is that someone else in Wuhan was working on the coronavirus and there was cross-contamination. Plausibly this is Wuhan Institute of Virology, who is known to work with coronaviruses. This is cool detective work, but it’s not clear what it’s supposed to prove. I think some lab leakers are using it to prove that WIV can do reverse genetics, but they admitted this already in a published paper so that’s not too helpful. I think others are using it to prove WIV had “secret viruses” in their catalogue, but the rice virus wasn’t secret, it was HKU4, which is common and which WIV has already published papers about. 1.6: DrJayChou’s 7 Arguments Once again, I cannot stress enough how much better a take you might have on this debate if you watch it. “The first known case predates the market outbreak by a month” - this is not the consensus position. I cannot say for sure what Dr. Chou means by this, but I suspect he’s referring to one of the many claims to this effect that Peter effectively debunked during the debate (Connor Reed, Mr. Chen, the 92 cases, Brazil, etc).
Inline links: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6178078/, https://www.bloomberg.com/news/articles/2023-07-18/us-suspends-wuhan-institute-funds-over-covid-stonewalling, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7022351/, https://www.biorxiv.org/content/10.1101/2023.02.12.528210v1.full, a published paper, has already published papers about, https://substackcdn.com/image/fetch/$s_!yA9U!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F467dd304-190a-4437-8920-d498c433dffb_1600x960.jpeg
China killed all the animals at the market after the pandemic started, without telling anybody which ones they were. We know more about them because a Chinese researcher had been documenting them for an unrelated project about tick-borne diseases. He sent his work to Western journals, and after a mysterious delay they eventually published them.
Inline links: a Chinese researcher
Your Review: Alpha School
June 27, 2025 · Original source
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