In other news, I have recently been selected as an Interact fellow. I am apparently now considered a “mission-driven technologist”. I do not know what that means, but I look forward to finding out.
Blogs
I have been saddened to learn that Pocket, the website I use to save articles for later and keep track of what I’ve read, is shutting down. For now, I’ll be switching to Instapaper.
A new collection of essays about Derek Parfit’s life is now available to pre-order in paperback. There is also an e-book, which you may be able to access for free already if you are affiliated with a university. I only read parts of volume two in this series. I also see that there is now an audiobook for Reasons and Persons, which is not the format I recommend…
A Stripe Press pop-up is coming up in Washington, D.C., on June 28th.
Second Act, Henry Oliver’s book about what late bloomers can tell you about reinventing your life, is now out in paperback. I gifted a signed copy to Julian Gough.
Something that’s striking about reading Chinese history is how much more common it is for numericexpressions to be the dominant way of referring to events, as with the May 4th movement or February 28 incident. There are many expressions primarily known by numeric epithets, like Hua Guofeng’s Two Whatevers, and important scientific and technological projects known by numeric codes (Project 596, Project 523).
But with the notable exception of 9/11, Americans will do just about anything to avoid naming something after a number. Exceptions like the Traitorous Eight prove the rule with how silly they sound. Gwern lists this cultural difference as among his most enduring unsolved mysteries:
[O]ne of the most striking aspects of Indian & Chinese political & religious rhetoric is the constant use of numerical epithets going beyond numerology, such as Gang of Four.
I must be missing something here: Isn’t this just because Chinese doesn’t have initial-based acronyms? There are several ways of abbreviating Chinese characters, but none of them are nearly as common as acronyms are in English.1Gang of Four is exactly the kind of grouping for which there would be a catchy acronym in Anglosphere political culture. Admittedly, this doesn’t solve the mystery for India, which is a different story.
I spent much of this month trying to learn some mathematics on Brilliant. I have now completed their tracks on group theory and linear algebra. Brilliant breaks my heart, in that they started with a fantastic product, which has steadily gotten worse with every update. They keep deleting courses I see no problem with, and the material they replace it with is much more introductory, and relatively well-covered elsewhere, like on Khan Academy. To access their group theory course, which might be their best, you have to scroll down to the archive, and I suspect it’s soon to be deleted.
I have high hopes for Math Academy, which I’ve just purchased a subscription for, and will be trying out this month.
Self-recommending: Dan Wang, Ezra Klein, and Derek Thompson chat.
I’m disappointed that more science podcasts… aren’t really about science. A recent discovery for me on this front is the 632nm podcast. They have recently been running a series about origins of life research; here they are talking to Anna Wang about her perspective coming into the field having been a soft matter physicist.
I was pleased to briefly chat with Jack Szostak at an origins of life conference last year (long story). It seems he’s recently been on the podcast circuit; here he is talking to Steve Levitt, and here he is on 632nm.
Jack Clark on AI diffusion scenarios. This is by no means meant as a dig, but it is quite striking that the highest educational attainment of the co-founder of Anthropic is a BA in Creative Writing from the University of East Anglia.
From Jane Street: much more than you ever wanted to know about how digital clocks are synchronised in modern finance. This episode was so incomprehensible that I gave up and read the transcript. I sent it to a quantitative trader friend, but he said it was too boring for him.
Buried within the episode is the revelation that Jane Street redid their whole tech stack for clock synchronisation because of a new European regulation requiring that trading firms be able to demonstrate that their clocks are accurate to within 100 microseconds of UTC. For over a year, I’ve been trying to convince someone to write an essay for Works in Progress or a similar outlet about the history of clock synchronisation in distributed computer systems. This was a Herculean project which is crucial to the functioning of many aspects of modern finance and the internet, and which relies upon monstrously complicated open-source databases and protocols that have been thanklessly maintained.
I suspect I would have been in exactly the target audience of a recent bootcamp in Berkeley for people who have no intention of working in finance, but have an abstract intellectual interest in quantitative trading. Sadly, it is so expensive that you would need to already work in finance to afford it.
Books
Bertrand Russell, Autobiography: 1872–1914. If you plan to read Bertrand Russell’s autobiography, it helps to know that it’s actually three books, which were written many years apart. Volume 1 is a masterpiece of literary memoir. Volume 2 is undeniably inferior. I’m told that volume 3 is a piece of shit.2
Russell doesn’t really discuss philosophy in his autobiography. It’s more about navel-gazing about his tortured marriages and personal life, and gossiping about his friends. Russell wrote a separate intellectual autobiography, My Philosophical Development, which, in my biased opinion is one of the few pieces of late Russell that is worth reading for non-specialists.
Whitehead was extraordinarily perfect as a teacher. He took a personal interest in those with whom he had to deal and knew both their strong and their weak points. He would elicit from a pupil the best of which a pupil was capable. He was never repressive, or sarcastic, or superior, or any of the things that inferior teachers like to be. I think that in all the abler young men with whom he came in contact he inspired, as he did in me, a very real and lasting affection.
And on the value of travel (p. 135):
Against my will, in the course of my travels, the belief that everything worth knowing was known at Cambridge gradually wore off. In this respect my travels were very useful to me.
Alex Ross, Wagnerism: Art and Politics in the Shadow of Music. A good book about Richard Wagner and his cultural influence. Pro tip: If you get the audiobook, it integrates excerpts from the most significant operas. The excerpts were much shorter than I had hoped, I assume because of copyright issues.
It seems that nothing is beyond the shadow of Wagner: I learned that the tune of ‘Here Comes the Bride’ is the bridal overture from the opera Lohengrin. This is quite ironic, given that in the plot of Lohengrin, the marriage is a disaster and the bride’s fate is tragic. As with so many things, the tune was popularised by the British Royal Family, who used it for the wedding of Princess Victoria and Frederick III (they were, after all, German).
Many people know the story – which has gotten progressively more embellished – about how there were riots in Paris at the premiere of Stravinsky’s Rite of Spring. But at Wagner concerts, riots and political violence were a regular occurrence! Perhaps the most striking thing about reading about the life of Wagner today is to appreciate the depth of admiration and disgust inspired by an art form that is, by today’s standards, quite a subtle acquired taste.
Maybe my favourite part was the (chapter 14) discussion about how Wagner came to be used in Apocalypse Now. Finally, Woody Allen’s thoughts on the matter: “I can’t listen to that much Wagner. I start to get the urge to conquer Poland.”
Ray Monk, Bertrand Russell: The Ghost of Madness 1921–1970. This is volume two of Monk’s biography of Bertrand Russell; I discussed part one in Links for March. I’ve never read a book where the contempt the author has for his chosen subject matter oozes off the page quite like this one. Monk started out much more sympathetic to Russell, and it really does seem like he regrets his choice to write the book.
Largely because of this contempt, these biographies are somewhat controversial. Here is a defence of Russell from A.C. Grayling that I find unconvincing. I am not qualified to opine about how much of Ralph Schoenman’s screeds that Russell agreed with, but Grayling’s implication that Russell played a substantial causal role in the Indo-Pakistan War and Cuban Missile Crisis I find ridiculous…
Despite Monk’s contempt for almost the entire life’s work of his subject matter, volume 1 won the book award from the Bertrand Russell Society, whose enthusiasm for the venerable Earl seems undimmed.
See page 191 for an error that would be shocking to make in the first edition of a book, let alone the paperback, in which Monk claims that John Maynard Keynes died in 1937.3 That came as a surprise to me, given that Keynes was one of Britain’s most influential economists and policy advisors during the Second World War.
(Notice, Mr Grayling, that I earn the right to be sarcastic by being a careful reader!)
Various, A Companion Guide to the Scottish National Gallery. When learning to appreciate visual art more, I’m not sure if there’s a substitute for living within walking distance of a gallery and going there multiple times a month. That was the position I used to be in with the Scottish National Gallery. By now, I reckon I could take a decent crack at giving a tour of any of the sections.4
One of the more significant early pieces of Netherlandish art in the collection, Gerard David’s (c. 1510) Three Legends of St Nicholas, has not solved for the equilibrium:
Better known to us as Santa Claus, the kindly patron of children, [Saint Nicholas] is next shown saving three girls from prostitution by giving money for their dowries.
Among my personal favourites in the Gallery is Hendrick Avercamp’s Winter Landscape, c. 1630:5
I saw a tweet recently about how to use LLMs to enhance your experience in museums and galleries: “An instant mini-essay on any painting, textile, or item with just a photo, no context necessary. Totally transforms the experience”. I wish I had the willpower to do this without getting distracted by notifications. Normally, I just end up checking WhatsApp.
It will be interesting to see if the network settings on the Daylight Computer allow it to connect to public WiFi. I like the idea of exploring new art where the only device I have with me is an e-ink reader that gives greyscale LLM output.
I didn’t realise that there are only slightly more than 30 surviving works by Vermeer. Edinburgh has his painting with perhaps his earliest, and only, Biblical subject matter, Christ in the House of Mary and Martha:6
Probably the painting in the Scottish National Gallery which has aroused the most speculation is The Skating Minister, since nobody knows for sure who it’s of:
Claude Shannon, A Mathematical Theory of Communication. I recently heard Grant Sanderson say that this paper is still the best introduction to information theory. It is 60 pages of dense electrical engineering, so I was putting a lot of trust in him to slog through it. I assigned it to my reading group, because it’s exactly the kind of challenging material that you get a lot more out of reading with friends. For reference, I was also consulting David MacKay’s Information Theory, Inference, and Learning Algorithms.
The form I read it in was this book from 1963, which contains the entirety of Shannon’s original paper, stapled together with a more introductory paper by Warren Weaver that explains the relevance to linguistics, philosophy, and the social sciences. While information theory has wide applicability, A Mathematical Theory of Communication is almost entirely concerned with signal processing theory and other relatively technical areas of engineering.
One of the key results here is essentially an existence proof of the possibility of error-correcting codes, via the noisy channel coding theorem.7 If I understood correctly, Shannon proved that, so long as the channel that a digital signal is being transmitted over has a greater capacity than the message itself (more entropy, which is the expected value of information), then, by using a sufficiently clever encoding system, you can make the rate of error as low as you want. He further showed that, subject to certain basic assumptions about what must be true of this concept of ‘information’, the unique mathematical form which can fulfil them is the concept of entropy from thermodynamics.
Nobody told me that Shannon’s central example to motivate how to formalise the definition of information is literally next-token prediction over a corpus of text. The man had already grokked many of the central philosophical and technical issues with large language models as early as 1948. Insanely visionary.
Papers
Shruti Rajagopalan, Alex Tabarrok, Premature Imitation and India’s Flailing State. The term ‘flailing state’ is due to Lant Pritchett. You might be under the impression from the extremely high level of government involvement in all aspects of Indians’ lives that an unusually high fraction of Indian society works in the public sector. But, scaled for population, only a fifth as many Indians as Americans work for the government.
This paper reminded me of a comment that Tyler Cowen made about Javier Milei: Whether you support having a Ministry of the Arts in the abstract is a different question from whether you support it when monthly inflation is 25%, or during a major fiscal crisis, or when 100 million of your citizens don’t have basic access to safe drinking water (as in India). People’s expectations about the appropriate level of government services seem to be almost completely unmoored from the level of actual administrative capacity those governments can realistically deploy.
This paper alleges that the (largely foreign-educated) Indian elites are particularly out of touch with the problems of the vast majority of the population, and that this results in systematic and predictable problems of policymaking. I had dinner with a smart and accomplished friend from Mumbai recently, where I disagreed with almost everything she said about Indians. She was surprised at the statistics I pulled up about how rare intercaste marriages are, and how they have not become notably more common since the 1970s.8 Similarly, 16% of Indians can speak “a little” in English, and only 4% of them are fluent.9 English is 44th (!) in the ranking of languages in India based on the number of native speakers.
Shruti and Alex give India’s maternity leave as an example of an Indian policy made more for the elites than the masses. Norway and Canada are the only countries in the world which have a longer period of paid maternity leave than India, a country with a GDP per capita of $2,500. This is despite the fact that only a very small fraction of Indian women work full-time in the formal sector. (The result, predictably, has been widespread noncompliance, and preemptive discrimination against women who might plausibly soon become pregnant.)
And while we’re on the topic of Indian labour law, the Industrial Disputes Act of 194710 mandates that any firm with more than 100 employees has to receive authorisation from the state government to fire any of its employees. If you have predicted that this means that India has a massively disproportionate number of companies with 99 employees, you would be correct.
The authors give many more examples. When you walk around Indian cities, they feel much more crowded than they should, given the population density. There is plenty of green and open space: It’s just hidden away inside private courtyards, which are necessary to comply with restrictive limitations on density. Mumbai has among the lowest floor space index of any major city. Many Indian cities also mandate parking minimums, in a country where only around 7% of families own a car. Time after time, Indian policies, if they were appropriate at all, would be appropriate only for a country at many times its level of economic development.
(In Inside the Mind of a Master Procrastinator, Tim Urban tells the story about how, two weeks out from his dissertation deadline, he procrastinated by spending a day on Google Earth to get a better feel for India. I’m not sure I’ve ever related to something so deeply.)
Simon DeDeo, Information Theory for Intelligent People. This paper comes from a presentation at the Santa Fe Institute’s Complex Systems Summer School, which a close friend describes as the best part of his entire PhD. It was a solid overview, although perhaps slightly eccentric in the applications. I’d definitely recommend it before tackling Shannon’s paper.
Philip Magness, Michael Makovi, The Mainstreaming of Marx: Measuring the Effect of the Russian Revolution on Karl Marx’s Influence. I was interested in this article in the context of evaluating the claim that the American political economist Henry George was equally, if not more, influential than Karl Marx on the political and intellectual left before the Russian Revolution. Scholars in other fields might be surprised by just how rarely economists think about Marx; Paul Samuelson famously called him “a minor post-Ricardian”. The authors claim that Marx was obscure in all academic disciplines before 1917, and would have remained so were it not for the Bolsheviks. Their attempt to do causal inference here involves a fancy econometric method called “synthetic control”, where you treat the Russian Revolution as a “treatment”, and you assume that whatever factors correlate with the rates at which Marx was cited beforethe treatment correlate in the same way afterthe treatment. You then use this to assemble a “synthetic Karl Marx” in the post-treatment time periods, which you compare to the real thing. The authors claim that this paper is the first time that synthetic control has been used on a corpus of text, which, if true, means that this is a wide open field for any political science PhD students reading…
Magness and Makovi find a substantial treatment effect (p.1536) in English and German, but not in French, which is fishy. The authors also don’t even think to mention until footnote 31 that they weren’t able to get bibliographic data on actual citations, so the outcome measure is the number of appearances of Karl Marx’s name in Google’s NGram database. So, at best, they’re measuring the cultural influence of Karl Marx in written material. Here is the key figure, page 1527:
I wonder if Scott Alexander cares enough about econometrics to write a short story in which “Synthetic Karl Marx” achieves consciousness and wakes up like Frankenstein’s monster.
There are some red flags here. Most of the organisations I could see that were involved in supporting this research are explicitly libertarian – with names like the “Hayek Fund for Scholars” – in a way that makes it wholly unsurprising that they reached a conclusion that makes Marx look bad. And when I see that Bryan Caplan is thanked in the acknowledgements of an economics paper, I honestly mildly update that the conclusion is false. The authors also don’t mention any of the known problems with the compositional changes in the NGram database. But in any case, it’s an interesting first paper in this literature. John Ganz’s allegation of “intellectual fraud” is absurd. For all its flaws, I think this is basically what it looks like when the system is working.
At a more foundational level, I’m uneasy with evaluating the content of a thinker’s ideas by looking at statistical trends in how much their work has been discussed over time. Suppose that one of my devoted blog readers overthrew the Intergalactic Federation in 2166 AD. It would be entirely unsurprising if more people read my blog afterwards. It would be quite odd if, instead of just reading my blog, future economists argued about me by applying sophisticated statistical methods to compare the actual world to a hypothetical world in which the new emperor’s coup did not succeed, or the one where he read Noah Smith’s blog instead. Those questions would be narrowly interesting as a matter of intellectual history, but it’s important not to overclaim:
While much of the discussion surrounding the bicentennial of Marx’s birth sought to differentiate consideration of his modern relevance from the totalitarian track record of twentieth-century communism, the elevation of Marx’s stature provided by the Russian Revolution illustrates that the two cannot be easily separated.
Not once, as far as I can tell, do the authors concede the trivial point that, if a thinker inspired a major change in many countries’ governments, it becomes more important to understand that thinker. I’d actually be more concerned if Marx’s citations didn’t increase after 1917. It would mean that academia wasn’t studying what was influential.
This was extremely far from the point, but the part of this paper I found most valuable was its exposition of the recent frontiers in multiple hypothesis testing. It seems that an important recent paper here is Wilson (2019), which argues that we should use the harmonic mean to average p-values. I should look into this.11
Alberto Cappi, The Cosmology of Edgar Allan Poe. What an amazing paper. Edgar Allan Poe was a keen amateur astronomer, and he wrote a remarkable poem in 1848 in which he proposed a novel theory of cosmology, which is regarded by astronomers as having been the first correct resolution to Olbers’ paradox. I will have more to say about this in a future post.
Anne Krueger, The Political Economy of the Rent-Seeking Society. A classic paper found via Shruti’s podcast. In this paper, Krueger builds a model in which it can be proven that the social welfare losses caused by rent-seeking over import licences are greater than the welfare losses from restrictions on trade in general. Bad social policies create a first-order inefficiency in their direct effects, and then a second-order inefficiency in how the spoils are divided between different interest groups. Many such cases.
Music
Richard Wagner, Orchestral Highlights from Operas. I enjoyed this recording from 1989 from the Slovak Philharmonic, which contained excerpts from The Flying Dutchman, Tannhauser, and Lohengrin. I am yet to hear any Wagner live, and my appreciation of a composer is usually pretty minimal up until that point.
Gustav Mahler, Mahler Symphony No. 1. After my exams ended, my first activity was to listen to the rest of the Mahler symphonies, while finishing a very good bottle of whiskey a friend left at my New Year’s party. Thankfully, I think if I had a genetic disposition toward alcoholism, it would have shown itself by now. Alas, none has been able to replicate the feeling of hearing Symphony No. 2 for me.
Alice Coltrane, Ptah the El Daoud. Yes, that is Alice, wife of John. This album predates her 26 years of making only Hindu devotional music, which I won’t claim to have listened to. She was one of the very few harpists in jazz – listen to it on Blue Nile, my favourite piece here. I have come to the conclusion that this is an A+ album, and will appeal to any fans of the Sun Ra Arkestra. And here is a mini-documentary about the church of John Coltrane in San Francisco.
老王樂隊, 吾十有五而志於學. My favourite find from Juan’s Chinese playlist. See here for some background; I think the album title means something like “stolen childhood”.
Films
Brian De Palma, Scarface. This is the first film I’ve seen by De Palma, and I didn’t even realise until after I finished it that it was a remake. Perhaps unsurprisingly, Al Pacino’s Cuban accent is… not good.
Why did nobody tell me that this film is literally about the Mariel boatlift? This was the historical episode in which 125,000 Cuban refugees showed up in Miami after Fidel Castro suddenly announced that Cubans wishing to leave the country could do so. It’s almost a rite of passage for a distinguished econometrician to write an analysis of the Mariel boatlift, which is the cleanest example we have of an exogenous shock to labour supply. It’s thus a great natural experiment to estimate the elasticity of labour demand (and thus, the effect of immigration on unemployment).12
In any case, I think I’ve just about satiated my appetite for gangster films for now; perhaps the next frontier will be Spaghetti Westerns.
Stanley Kubrick, Full Metal Jacket. I agree with what seems like the critical consensus that the film loses a lot of momentum after the iconic boot camp scenes. Those scenes have been parodied so much that it’s not an easy film to watch for the first time in 2025. There has also been quite a disturbing afterlife to the Vietnamese prostitute scene, which I was unaware of.
Kenneth Clark, Civilisation. Note this is Kenneth Clark, the art historian, not John Major’s Chancellor of the Exchequer. As soon as my exams ended, I hosted a dinner party for my friends to start watching this documentary series, which is about the history of Western art and culture. This was per the recommendation of Jamie Rumbelow, who told me that Clark did for the arts what Carl Sagan did for astronomy in Cosmos.
I didn’t know that Raphael has the reputation for difficulty that he apparently has. I also haven’t been paying nearly enough attention to Bernini. Before watching this series, I didn’t know anything about the explosion of creativity in Rome around 1620–1640 (the beginning of the Baroque period). Clark claims that “fifty years ago” (which I suppose means ~1919), Mozart was taught as a straightforward Rococo composer, and not the era-transcending genius he became known as later. It seems that a lot of Mozart is relatively formulaic, such that he repays having a good guide; next, I will be looking to the Jan Swafford biography.13
Watching all these BBC broadcasts from the 1960s has made me think that The Simpsons were on to something with the Big Book of British Smiles.
Finally, from YouTube, a solid concert: The Emmet Cohen Trio live in Shanghai. Elsewhere, Grant Sanderson explains Grover’s algorithm, and dispels the myth that quantum computing has anything to do with “trying all the combinations in parallel”. As Scott Aaronson says, this video is a masterpiece. Note there is also a later correction. Exponential speedups from quantum algorithms are the exception, not the norm!
Thanks for reading Sam Enright's Newsletter! Subscribe for free to receive new posts and support my work.
Keynes said of Russell that “Bertie held two ludicrously incompatible beliefs: on the one hand he believed that all the problems of the world stemmed from conducting human affairs in a most irrational way; on the other that the solution was simple, since all we had to do was to behave rationally.” See page 177 of Monk’s book.
Not Mary as in the mother of Jesus. Or Mary Magdalane. Mary as in Luke 10:38–42. I actually found one of the more difficult elements of reading the Gospels was keeping track of the excessive number of people called Mary.
There are lots of classification issues here that make this a non-trivial claim to evaluate. Besides, in English the word ‘caste’ merges together two different concepts, Jāti and Varna, the relationship between which it itself non-trivial and differs based on the state. My o3 conversation contains some helpful links about this.
The original paper has 239 citations, but only 4 citations for the quickly-issued correction – sad! The only reason why I can even remember what a “harmonic mean” is because of my maths olympiad days, when knowing which of the different ways of “averaging” a set of numbers are weakly greater than one another is helpful on the easy number theory problems (which were the only ones I knew how to do).
The main conclusion of these papers, by the way, is that immigration doesn’t increase unemployment very much, if at all, although George Borjas is a dissenter.
Maybe the Chinese case is explained by the lack of initialisation, but the puzzle goes a bit wider than that. For example: why does Hebrew lean into acronyms _so much more_ than any other language (that I know of)? Basically half of Rabbinic literature is either asserting that some word should really be interpreted as an acronym, or else creating a new acronym. They do this to the point that almost every possible combination of Hebrew letters has at least two possible acronym meanings, almost certainly many more [https://en.wikipedia.org/wiki/List_of_Hebrew_abbreviations]. If your claim is "logographic languages use numbers, alphabetical languages use acronyms", the question isn't just why Indian languages use numbers, but also why Hebrew uses acronyms *way more*.
Chris Olah, another Anthropic founder, famously does not have an undergraduate degree.
Maybe the Chinese case is explained by the lack of initialisation, but the puzzle goes a bit wider than that. For example: why does Hebrew lean into acronyms _so much more_ than any other language (that I know of)? Basically half of Rabbinic literature is either asserting that some word should really be interpreted as an acronym, or else creating a new acronym. They do this to the point that almost every possible combination of Hebrew letters has at least two possible acronym meanings, almost certainly many more [https://en.wikipedia.org/wiki/List_of_Hebrew_abbreviations]. If your claim is "logographic languages use numbers, alphabetical languages use acronyms", the question isn't just why Indian languages use numbers, but also why Hebrew uses acronyms *way more*.