Links & What I’ve Been Reading, October

Media recommendations

Here are the highlights of the most interesting things I’ve read, watched, and listened to in the last month. But first, two announcements:

First, I will be in California this weekend, and I’m hosting a meetup at the Long Now Foundation in San Francisco on October 20th at 3pm. This will be your chance to meet the Irish mafia who secretly control Silicon Valley and all of global politics. You can join the waitlist to RSVP at the link.

Second, thank you to everybody who has pledged money to my blog. I have no immediate plans to put anything I’m involved with behind a paywall. My favourite part is seeing the messages people leave for why they want to pledge, which read like they’re from a fortune cookie: 

If you want to support me, you can donate to Progress Ireland.

Blogs

From Trevor Chow: Three Kuhnian revolutions in ML training. Four years ago, everybody knew that when you’re training a machine learning model, you should spend more of your computation (‘compute’) on parameters than data. Two years ago, the paradigm shifted and suddenly everybody believed that you should split it equally across parameters and data. And last year, it became widely accepted that you should be spending orders of magnitude more on data. Why? What’s going on here? How can the modus operandi of an entire field completely reverse so quickly? As per usual, the philosophy of science of Thomas Kuhn (he of ‘paradigm shift’ fame) is being abused – but I’ll forgive it. And: some words about GPT-o1.

Seán Keyes on the extraordinary cost savings possible in Anglo countries if they stopped relying so extensively on contractors and consultants to procure large infrastructure projects.

A different Seán on how a street vote policy could achieve much greater housing density and affordability in Dublin. And: the surprising amount of new housing supply enabled by allowing people to build in their back gardens.

Luke Fehily on how ‘metascience’ can be used to make research funding in small countries competitive.

Anson Yu on sparkly people and how to find them. Editor’s note: Anson is herself a sparkly person.

“Why spend one centavo on cosmology when little girls are crouching in subway stairwells, begging?” 

If you’re anything like me, you often come across extremely talented engineers with blogs about their eccentric side interests. The latest ones I’ve found are Zhengdong Wang and Jannik Schilling. From Zhengdong, here are some fun facts about shipping containers: parts one and two. 

The road from serfdom, a long article about the end of serfdom in the Russian Empire and how Pyotr Stoylpin achieved his agricultural reforms. I felt that the comparison with modern political change was forced and underdeveloped, but I learned a lot from reading in any case.

A review of David Cox’s Galois theory textbook. Évariste Galois was a French mathematician who pioneered almost the entire field of abstract algebra in letter (!) to his friend (!!) after being expelled from university (!!!). He was involved in overthrowing King Charles X during the 1830 Revolution, and then died in a duel over a woman when he was 20 years old. As soon as Lin-Manuel Miranda finds out about Galois, there will be a musical about Galois. 

(Wait! There could also be a musical called ‘Hamilton’ about the nearly equally interesting life of Irish mathematician William Rowan Hamilton and his discovery of quaternions. Now I can’t decide which spoof musical about 19th-century contributions to mathematics would be better.)   

Logistic regression is not fucked. Unfortunately, this gem of a data science blog seems to have been hacked by an Indonesian gambling website (?), so you have to access the link via Internet Archive. The bread and butter of statistical inference is regression, and sometimes you want a regression to predict the outcome of a binary variable. Let’s say you have a dataset of the characteristics of people on the Titanic, and you want a model to output the probability that a hypothetical person with certain characteristics would have drowned. If you use linear regression, it will sometimes predict values that aren’t between 0 and 1. So instead you can transform the data with a function that maps all inputs from -∞ to +∞ to outputs between 0 and 1. One way to do this is log odds, i.e. the logarithm of the ratio of the probability that something will happen to the probability that it won’t. If you do it that way, you get logistic regression. There’s been a debate going on in recent years about whether these forms of regression are even worth it, or if it’s fine to just estimate things linearly (with a hacky solution like declaring all negative values are 0 by fiat). This post was a nice argument that logistic regression is one of the few things in empirical social science which doesn’t have major problems.

Scott Sumner has stopped blogging at The Money Illusion and has now moved to Substack. Here he is analysing trends in 16th-century Italian art. His new crop of film reviews has been released, all of which are now easily viewable online. I say again!

Paul Graham: Why are so few of the best founders mean?

I’ve been enjoying messing around in this database of every Anglo-Saxon whose name we know. 

Brief reviews of several mathematics textbooks. I know this doesn’t sound like it would be incredibly funny, but it somehow is. And here is Raymond Smullyan, the author of the continuum hypothesis textbook mentioned, speaking about logic puzzles with Johnny Carson in 1982.   

(Parts of) modern economics research is basically a field of software engineering. And yet, the software norms are closer to 1970s academia than to modern startups. Maxwell Tabarrok on how programming norms in economics should improve. 

Power analysis is the practice of figuring out the statistical power of a proposed study, which is the probability that your methodology would detect an effect, assuming the effect is real. Sadly, many scientists eschew the practice of formal power analysis and instead rely on rules of thumb. Here’s an argument why that’s bad. 

While on the topic of structural problems in science academia, here is the syllabus for ‘Everything is Fucked’, Sanjay Srivastava’s seminar about replication and problems with statistical inference. Two more from his blog: science is more interesting when it’s true, and what is it like to work inside an infant cognition lab?  

Niko McCarthy’s list of underrated science books. To this list, I’d add The Sage of Science, Andrew Brown’s biography of the X-ray crystallographer J.D. Bernal. It’s the authoritative biography of the man who James Watson considered to be the father of molecular biology, and yet I’ve never met anyone else who’s read it. 

Isaak Freeman on how he learned to speak Chinese in a year. As if studying for an undergraduate degree and an MIT PhD at the same time wasn’t difficult enough, Isaak decided to also do his PhD in Mandarin. He is a year younger than me – there truly is no hope. A few months ago, I unsuccessfully tried to get Isaak to order for us in Chinese at a Sichuanese restaurant.

Biology and entrepreneurship links roundup from Jose Luis Ricón. 

Fertility links roundup from Works in Progress. You may have seen the terrifying statistic that it’s gotten substantially more dangerous for women to give birth in the US over the last 20 years. Saloni Dattani explains why this is false and how the whole phenomenon was caused by our old friend, measurement error.      

Gwern asks: Why are there so few Matt Levines?

Stupid party ideas for your stupid friends. Last year for my birthday, I ran a ‘puzzle party’ in which we converted two of the rooms in my house into lateral thinking puzzles that my friends had to solve in teams of six or eight. Perhaps at some point, I’ll write up some notes and advice about how to throw memorable parties. 

Why did Russians excel in mathematics and other technical fields – and have so much fun while doing it? One possible answer is the practice, going back to the Russian Empire, of ‘mathematical circles’, in which bright children continually grapple with problems above their current ability to solve completely.

Book review of ‘Science in Traditional China’, Joseph Needham’s esteemed tome about the decline of Chinese science and inventiveness. There is a widespread urban legend that gunpowder was invented in China, but for the first few centuries was only used for fireworks and had no military applications. This is completely false. I was also surprised to learn that, while English has few words of Chinese origin, ‘ketchup’ is (probably) one of them. 

The basics of Chinese medieval warfare. The equivalent of the ‘dark ages’ in China was between ~220 and ~620 AD, i.e. after the Han and before the Tang Dynasties. By all accounts, it seems to have been a Mad Max-style anarchic hellscape.

Dynomight against dystopian interpretations of listening to audiobooks at 2x speed.  

How Avital Balwit uses Claude. Avital is now the chief of staff at Anthropic. I’ve recently switched over almost entirely from GPT-4o to Claude. Sometimes I compare different LLMs’ responses to the same prompt, especially if I’m worried about hallucinations. Does anybody know if there’s research about how correlated hallucinations are across different models? If all the frontier LLMs tend to give incorrect answers to the same sort of prompt, then presumably this strategy of getting models to check one another is of more limited value. 

Podcasts

Rasheed Griffith interviews Henry Oliver about V.S. Naipaul, Dublin as a literary city, and how to get into non-Western literature. The Fitzwilliam receives two indirect mentions, first as the source of Henry’s essay about James Joyce, and second, why Rasheed was writing an essay about Irish-Caribbean connections. Henry Oliver now has his own podcast, although I haven’t listened yet for the simple reason that I’m just now realising that it’s available on Spotify and not just the Substack app.   

Dwarkesh Patel interviews Daniel Yergin about his book The Prize, a global history of the oil industry. One of my new favourite podcast episodes ever. At some point, I may pen something about Winston Churchill’s attitude toward technology in the 1910s and 1920s. As First Lord of the Admiralty, he was particularly early to recognise that the wars of the 20th century would be about the control and utilisation of oil. He also wrote an essay about lab-grown meat. And: a free-wheeling conversation about the tacit knowledge of the semiconductor industry.  

My friend Julian Gough was interviewed about writing the Minecraft End Poem, what you learn from writing children’s books, and theories of cosmology. 

The 99% Invisible series about the Power Broker has continued.

In Our Time, on BBC Radio 4, is one of the best things the BBC produces. Here is the recent rebroadcast of their episode about the philosopher Elizabeth Anscombe. I think that we should have put Melvyn Bragg on the Voyager Golden Disc. 

Venki Ramakrishnan surveys the current state of life extension research. 

Very Bad Wizards discuss Borges once again. 

Yasheng Huang on the legacies of Imperial China for today, parts one and two.

The Empire podcast on the Darien scheme and the 1707 Act of Union.

This month I listened to the back catalogue of the China Book Review podcast, a companion to the China Books Review. My favourite episodes were about remembering Tiananmen Square, and the influence of the sociologist Chan Hansheng. 

I recently learned about a group of leftist Filipino students who went on an exchange to the People’s Republic of China in the early 1970s. It wasn’t safe for them to return after Ferdinand Marcos declared martial law, and eventually their passports expired, leaving them stranded during the Cultural Revolution. They did not speak Chinese. One of those students, Jaime FlorCruz, is now the Philippines’ ambassador to China, and I listened to this very interesting (gated) interview with him. 

Tom Tugendhat on political reform in the UK, and what’s special about Kent. I mostly liked this, but his answer about the social value of private schooling was weak and evasive. 

From the Institute for Progress: Metascience 101, an overview of recent thinking about how to improve scientific institutions, with a great lineup including Heidi Williams and Alexander Berger. Self-recommending! 

Tyler Cowen and Alex Tabarrok talk about who will, and who should, win the Nobel Prize in Economics this year. And: Tyler and Alex on the causes of the 1970s inflation.  

Papers

Ronald Lee, The Demographic Transition: Three Centuries of Fundamental Change. Pretty standard review paper about long-run demographic changes and population projections. Time to be cranky again: Can somebody please explain why the American Economic Review clutters all their PDFs by appending 20 pages of ‘this article has been cited by…’? Who needs that kind of information in the actual paper? Why isn’t just a searchable list on the website? Academia has a ridiculous skeuomorphic insistence that a “paper” must emulate a self-contained physical paper.

Karen Eggleston, Victor Fuchs, The New Demographic Transition: The Gains in Life Expectancy Now Realized Late in Life. The bottom line of this paper is that, while we (well, most of us) are still getting life expectancy gains, they are coming from reductions in mortality late in life, and mortality rates at younger ages are relatively stagnant. This paper made it click for me just how much favourable demographics aided the East Asian miracles. One estimate is that 15% of Chinese growth in output per capita can be explained just by demographics, and another estimate is that between a quarter and a third of growth rates in the East Asian Tigers were ‘demographic dividends’.

Melissa Kearney, Phillip Levine, Luke Pardue, The Puzzle of Falling US Birth Rates since the Great Recession. I previously blogged about Levine and Kearney’s paper about estimating the causal effects of Sesame Street. This paper can be construed as a minor win for those who argue that culture trumps policy when it comes to fertility, as the authors “are unable to identify a strong link between any specific policies or economic factors and the declining birth rates.” I suppose that’s why it’s a puzzle. Kearney and Levine have another paper in which they claim to show that the introduction of the TV series 16 and Pregnant caused a large and measurable (!) decline in teen pregnancy. Something interesting about the American case is that, in a sense, the null hypothesis should not be constant birthrates; if birthrates had stayed the same within each race, then fertility would be rising (since the population now has a larger share of ethnic minorities with higher birthrates). It’s funny to see Lyman Stone and Ross Douthat be cited in a top economics journal. 

Tiloka de Silva, Silvana Tenreyro, Population Control Policies and Fertility Convergence. The main thesis of this paper is that explicit efforts to limit fertility rates were highly successful. The authors further argue that this is the primary reason why East Asia has lower fertility than other places at similar levels of economic development. Similarly, due to the HIV pandemic, family planning efforts in Africa were directed away from population control policies, and they argue that this is why Africa has unusually high fertility for its level of development.¹ China’s one-child policy is the most famous example of a population control policy, but there were all kinds of economic and social incentives – some more coercive than others – and education campaigns to encourage women to have fewer babies (some of it funded by Western NGOs!). Even today, there are more governments with explicit goals to lower fertility than to raise it. This paper was also the first time I’ve seen an estimate for the total fertility rate of the world as a whole, which turns out to be about 2.3 births per woman.

Indian family planning poster which reads, in Marathi, “…two to three children… are enough”

I read all these demography papers for a reading group that a friend hosts for undergraduates. He and I were commenting that we’d both love to have a picture book of family planning posters throughout history. The Chinese posters in the ‘Socialist realism’ art style are amazing.

Matthew Gentzkow, Jesse Shapiro, Code and Data for the Social Sciences: A Practitioner’s Guide. Some advice on writing ‘clean code’ for social science research projects. Found via Maxwell’s blog linked above.

Michael Huemer, Correct Resolution of the Twin Paradox. You may have heard of the twin paradox, a ‘paradox’ in which one twin leaves the Earth on a spaceship while another stays home. In special relativity, motion is always relative, and observers moving at a constant velocity can think of themselves as being at rest. And also, time slows down as you move closer to the speed of light (time dilation). So, Earth Twin will think that Space Twin aged more slowly, and so Space Twin will look younger if the twins reunite. But according to Space Twin, Earth Twin was the one moving, so Earth Twin will look younger if they reunite. Which of them is right?

Many people give the incorrect answer that Space Twin has to be the one who aged slower, because it was she who had to accelerate in order to reach a high speed. In fact, acceleration and general relativity have nothing to do with the resolution to the paradox (the incorrect answer sometimes appears in secondary school physics classes, and I’m almost certain I’ve seen it on YouTube videos before). The correct answer is that, if the twins never meet again, there’s nothing contradictory about them both being younger than each other (I know, weird). And if they do meet again, then it’s easier to see why there’s no contradiction on a spacetime diagram.

Roy Sorensen, Yablo’s Paradox and Kindred Infinite Liars. The philosopher Stephen Yablo is known for an ingenious reformulation of the Liar’s paradox (‘this sentence is a lie’) in which the paradox has nothing to do with the idea of self-reference. That would mean that attempted solutions to such paradoxes which restrict when self-referential statements can be meaningful won’t work. Ultimately, it doesn’t seem very plausible that the correct resolution to a simple semantic paradox is as elaborate as what is described here, but it’s a nicely written paper in any case.

Scott Aaronson, Why Philosophers Should Care About Computational Complexity. This is one of the most mind-expanding papers I’ve ever read. Computational complexity is about how efficiently algorithms can be computed as a function of the inputs. To take just one example he discusses, under what conditions will it be computationally tractable for rational agents to find a Nash equilibrium? It could be the case that, while economics has lots of nice theorems about the existence of certain equilibria in markets, actually finding those equilibria is intractable. And yet, most economists and philosophers barely even know what computational complexity theory is! Aaronson also ties in complexity theory with the strong AI debate, the Chinese room argument, and presents the most original perspective on Hume’s problem of induction I’ve ever seen. I also learned from this paper that Kurt Gödel doubted Darwinian evolution because he thought that the Earth had not existed for long enough for an ‘evolution algorithm’ to produce such a complex output!

I loved this paper so much that I’m running a reading group to closely read and discuss it. It’ll be on December 14th in Edinburgh – if you’d like to join, you can message me.

Adam Rieger, Paradox, ZF, and the Axiom of Foundation. The foundations of mathematics are formulated in terms of set theory. Until the 20th century, people believed the ‘naive comprehension axiom’: that, for any property, there is a set (maybe an infinitely large set) of all the things which have that property. This runs into problems of self-reference, like if you want to construct the set of all sets that don’t contain themselves (Russell’s paradox). There are various ways to modify the dominant (‘Zermelo-Fraenkel’) formulation of set theory to account for these problems. This paper is about one particular approach, which is to add an ‘axiom of foundation’, which says that for any set A, A must contain some element which is disjoint (i.e. has some elements which differ) from A.

Thomas Kelleher, Manual on Export Free Zones (no easily accessible pdf, sorry). This was a publication from the United Nations Industrial Development Organisation in 1976 which some academics think inspired the legislation for and implementation of several special economic zones around the world. It draws on the Irish experience to argue for free trade zones as a viable component in developing countries’ economic policies. I’m now 98% sure that this document was studied by several people in Deng Xiaoping’s staff in either 1979 or 1980, and that it influenced their thinking while on a UN-funded ‘study tour’ of how they might liberalise the Chinese economy.² This is not written down anywhere: I heard it from the guy who heard it from Jiang Zemin (Deng’s successor as paramount leader of China). So there you have it, breaking news on the Sam Enright blog. 

Paul Benacerraf, Tasks, Super-Tasks, and the Modern Eleatics. This paper is in response to a famous 1954 article by J.F. Thomson. That article is the origin of the idea of ‘Thomson’s lamp’. If a lamp starts by being turned on, and then it’s switched off after half an hour, then back on again after 15 minutes, then off again after 7.5 minutes, and so on… then will the lamp be turned on or off after a full hour? Such a situation is an example of a ‘supertask’, about which Vsauce has a pretty good video. If you’re looking for an (opinionated) overview of paradoxes of infinity, instead of this paper I’d recommend Michael Huemer’s Approaching Infinity. PS: ‘Eleatics’ is an (unnecessarily obscure) reference to Zeno of Elea (he of ‘Zeno’s paradox’).     

Lucy Page, Rohini Pande, Ending Global Poverty: Why Money Isn’t Enough. Here’s a trend in global poverty you may not know about: In 1987, 90% of the world’s poor lived in low-income countries. By 2013, over 60% of the world’s poor lived in middle-income countries, and almost half of the extremely poor lived in middle-income countries. Extreme poverty has come down a lot, especially in relative terms, but the poverty that remains has to a significant extent shifted to countries which are themselves middle-income. Before reading this paper, I didn’t realise the extent to which the extremely poor are now split between increasingly unequal high-growth countries, and failed states – an unusual situation! Page 179 has some useful statistics about how unequal growth in India has been; unsurprisingly, lower castes now represent a higher fraction of the poor than they did in the 1980s. This paper also features a cameo from an old friend of the effective altruists, the PlayPump International merry-go-round. Here is a key graph: on net, the extremely poor of Indonesia donate to international development assistance!  

The y-axis is dollars

Rema Hanna, Benjamin Olken, Universal Basic Incomes versus Targeted Transfers: Anti-Poverty Programs in Developing Countries. You probably know about GiveDirectly, a charity which sends donors’ money directly to people in low-income countries via an app on their phone. The direct cash transfer approach has shown itself to be highly effective, and generally in development, it’s relatively challenging to beat cash transfers. This paper is a nice overview of the recent literature on whether those cash transfers should be universal or targeted within specific groups. The authors do some back-of-the-envelope calculations and build a basic model to calculate the optimal rates of false positives and false negatives under different assumptions. The result comes out favourably toward targetting as many social programmes as you can.

Music

Gustav Mahler, 2nd Symphony. This symphony is called ‘Resurrection’, and I gather that it is (among other things) influenced by Mahler’s pragmatic conversion from Judaism to Roman Catholicism. Mahler is one of the more experimental of the ‘greats’; this symphony has five movements, of quite different lengths, breaking with the convention of four. Two weeks ago, I listened to it being performed by the Royal National Symphony Orchestra. I suppose Mahler is not the best composer to start with if you’re new to classical music, but the performance was excellent and it was the second-best classical concert I’ve been to (after Jack Gibbons). I’m sorely tempted to buy tickets to see the same orchestra perform a cultural contribution of similar stature, i.e. the Muppet Christmas Carol soundtrack. 

Keith Jarrett, Facing You. Solo Jarrett which was previously unknown to me. If you want more jazz recommendations you should read Andrew Batson’s blog.   

Ngô Hồng Quang, Nhìn Lại. Interesting Vietnamese music. Hat-tip Zhengdong Wang. At some point, I should learn how Vietnamese transliteration works; is there a specific reason why Vietnamese makes considerably more use of diacritics than the Romanised forms of other Asian languages?  

Jupiter & Okwess, Na Kozonga, especially ‘You Sold Me a Dream’. Wikipedia describes this genre as ‘Bofenia rock’, a Congolese variant of Afropop. This led me down a rabbit hole about the history of African music; the page about Zaire ’74 is well worth a read.  

Cleo Sol, Gold. Smooth UK R&B from a half-Serbian half-Jamaican singer. Not really my kind of music – it was nice to write to, I suppose?

Books

Stefano Sandrone, Nobel Life: Conversations with 24 Nobel Laureates. It’s entirely a coincidence that I finished this book during Nobel announcement week. Mostly this is just harmless intellectual ‘popcorn’ of various scientists reflecting on their careers. However, I only gave it two stars because the book never explains what the ‘Lindau Meetings’ are, despite being the central premise of the book, the references are formatted annoyingly,  and the questions are not well-researched. We should not stand for this in the age of Dwarkesh Patel! I didn’t know that Bob Solow not only fought in World War II, but was among the first Allied troops to reach Rome. 

Do I have any opinions about the Nobel announcements? Not especially. If you were starting a new prize from scratch, you’d presumably carve up the disciplines in a different way, such that fewer placements of individual prizes would look silly. Geoffrey Hinton has been giving us some funny moments: 

John Stillwell, Elements of Mathematics from Euclid to Gödel. I loved this book. This is much more philosophically and historically rich than a typical popular maths book. My only (minor) complaint is that the author is obsessed with what counts as ‘elementary’ mathematics, which came across to me as arbitrary and uninteresting. I’ve mentioned before my difficulty in finding books in the missing middle between popular maths and textbooks, and it seems that the Princeton Science Library may be just what I was looking for.

Liz Rideal, How to Read a Painting. For some time, I’ve wanted to read a book along the lines of ‘the 100 most famous paintings, explained’. This didn’t fulfil my hopes: the pages are too small to appreciate the artwork. Getting absorbed in one artist at a time and reading biographies and such seems to be a better way of cultivating appreciation, at least for me. Next on my art history reading list is Sue Prideuax’s new biography of Paul Gaugin. I saw the author speak recently, and she was great. A new painting I discovered from Rideal’s book and really like is Paul Cézanne’s Mont Sainte-Victoire & the Viaduct of the Arc River Valley: 

Films

David Litchfield, One Hand Clapping. This film consists of digitally restored footage of Paul McCartney recording sessions from the 1970s with his band Wings in Abbey Road Studios. It was screened briefly in art house cinemas across the UK two weeks ago. There are so many defects in the film that the idea that it’s a ‘4k restoration’ is basically a scam, but you’re not primarily going for the visuals. All my friends were busy when this was showing near me, which was in the middle of the day. So I went alone, which was the most unemployed I’ve ever felt.   

Ken Burns, Vietnam War. I’ve been watching this series with a friend, and we’ve still only made it to episode five (of ten). I’ll be honest, this series is too long, and I suspect that the abridged edition which aired on the BBC a few years ago is superior. The filmmaking doesn’t distinguish between which setup is for genuinely consequential figures in the war whose names you should remember, and which is for generic soldiers being used for illustrative purposes. Almost the equivalent of an entire episode is spent on “Mogie” Crocker, an American soldier who died in Vietnam and… is not famous or historically significant. The 320,000 South Koreans who fought in Vietnam scarcely get a single mention. Swati Sharma in Vox writes: “I can’t imagine watching more than an episode or two at a time. There is something dark and murky and repetitive about the whole enterprise… [that] turns the film itself into a sort of metaphor for the Vietnam War.” Indeed. 

Finally, from YouTube, I’ve been enjoying these real analysis video lectures. And: the beauty of the Banach fixed point theorem. 

1. I didn’t even realise this was true, but it is. To take one example, Botswana’s GDP per capita (nominal) is ~$8,000, and the TFR is 2.8 births per woman. India’s GDP per capita is ~$2,400, and the TFR is 2.08 births per woman (below replacement!). While fertility rates generally decline as countries get richer, they have declined much more slowly in Africa than elsewhere.

2. I’ve been unable to find any Chinese translation in the UN archives, so I’m not sure how they read it. It’s worth noting that Jiang Zemin had much greater proficiency in English than almost any of the Chinese leadership of his generation.