Links for March

Interest rates, Bradford, Peter Pan

Here are the most interesting things I read, listened to, and watched in the last month. But first, some announcements:

1. I will be in London March 31st and April 1st. I will then be in London again from April 24th to 26th. If you would like to meet up, shoot me an email or send me a DM.

2. If you are in New York, please come to the Fitzwilliam meetup on April 16th. You can RSVP here. +1s welcome.

3. I anticipate the next Fitzwilliam reading group to be particularly strong. We will be doing a deep dive into the social and economic consequences of railroad construction, focusing on Ireland, British India, and America. It will be in Dublin on April 11th, with special guest Alan Fernihough. Send me an email if interested.

4. Similarly, the next Fitzwilliam maths circle should be fun: we will be checking whether we still remember how to solve integrals. That will be on April 6th.

5. A reminder that, if you find what I do valuable, please consider upgrading to a paid subscription. <3

6. There is also an ongoing debate about the legacy of Tyler Cowen, prompted by an economist criticising him. I am not much of a fan of tweeting, so I will add my two cents here. I’ve talked to Tyler most weeks since I was a teenager. He’s never taken more than a few hours to respond to one of my emails or texts, or failed to make time for dinner when we’ve been in the same city. He’s provided feedback on many of my blog posts, essays, and papers. He has given me three Emergent Ventures grants, had me on his podcast, introduced me to many of my closest friends and mentors, and is responsible for a large fraction of my blog subscribers through consistently featuring my writing on Marginal Revolution. The amount I personally owe to him is genuinely unfathomable.

I am not even unusual for that. I have been to house parties at which literally more people than not owe a major career pivot or beneficial personal decision to him. He hosts two of my favourite podcasts, writes the most successful economics blog of all time, has written over a dozen popular books, a major textbook, hosts an online education platform, and has distributed tens of millions of dollars in philanthropy. He clearly has some strange opinions, which I have no desire to pretend that I agree with. But if you don’t recognise that he must be among the most impressive and personally generous human beings currently on the planet, you are unserious.

Blogs and short links

The proceedings for the Fitzwilliam conference about Milton Friedman have now been posted. Featuring Rebecca Lowe, Robin Hanson, Agnes Callard, Anup Malani, Sebastian Garren, and some other faces you may recognise. I had a lot of fun writing this, and am happy with how it turned out.

For Progress Ireland, I wrote about the economic incidence of corporation tax:

[W]here Ireland is an extreme outlier internationally is in how much of that tax is paid by a relatively small number of American multinational companies. Only 11 percent of Irish corporation tax is paid by Irish companies. The distribution has become so skewed to the point of boggling the mind: in 2024, Apple and Microsoft combined paid 40 percent of all corporate tax. Ten companies pay €0.60 out of every euro of corporation tax.

This was a month of many linkposts. My Links for January were abridged and reposted on EconLog. There were also José Luis Rícon’s links. We also had Dan Frank’s paean to linkposts, Lauren Gilbert’s weekly links, and Slime Mold Time Mold’s links for February.

In honour of St Patrick’s Day: How Irish cooks became so legendarily awful.

Contra educators who believe that constant repetition turns students into mindless drones. And, the latest entry in “memorisation is underrated”: we should all be learning how to recite more poems by heart.

Best of Twitter: “There’s an old saying in Schelling – I know it’s in Hegel”. I can confirm that this is true. Also: my latest power couple discovery.

Your periodic reminder of how batshit insane the Taiping Rebellion was.

The Abundance and Growth Fund at Coefficient Giving now has its own blog, where they update you about what they’ve been reading. Along with unbanning supersonic flight, is this the second thing Trump II has done which is actually good?

[T]he Trump administration is suspending the Jones Act.

One of the wisest fellows around is the great Witold Więcek. Here is his February update:

Throughout my “career” as a “statistician”, 13 years and counting now (but how much longer?), I’ve always been great at stopping myself from doing useful work. At first, I worried that I didn’t know enough yet to tackle interesting problems---until I’ve started feeling that I forgot too much to do “real” statistics. With LLMs that barrier is now gone and I’ve been finding them very useful. I have just enough context and experience to pose good questions and understand the explanations.

(BTW, I am surprised by how little students and my peers seem to use them. I am usually, willingly, cast in the role of the nay-sayer. So what’s happening? Are they using them surreptitiously? Or else, why do I get more utility than others?)

But my work also got more boring now. Difficulty dial acts on both the good-challenging and bad-challenging bits and in some situations I am reduced to being a fleshy interface for a vast inhuman entity---parse the signals, feed into the machine, see what it spews out, translate.

So, obviously I decided to make this situation even worse and dip my toes into THE AGENTS this month (starting with the OpenAI one). In case you haven’t encountered them, these are the ~latest craze in the LLM world.

Yes, just like with chatbots, you just describe what you’d like, in words, and it gets coded. But it’s not just coding programs. You can do (some parts of) academic research or you can just make small, fun ideas come to life. I recently met a girl who vibe coded a Chinese medicine app that took photo of your tongue and told you seven things that were fucked up about your bladder.

Ultimately, however, my problem---because obviously I wouldn’t bother to write this to just conclude that they’re alright, would I---is that these tools are designed for people who like manipulating mental symbols in a certain way, you know, the screen-starers. Obviously this is a ridiculous complaint, not least because I am one of them... but as I get older, screen-staring part of my brain feels like the one I want to be visiting least often. And I think it was no coincidence that I had most fun playing with these tools when my mood was lowest.

In fact, they are addictive as hell, like a video game can feel. Everyone keeps reporting this. They dial difficulty down so much, that things get a bit muddy. People who talk about these models the most often seem to me maniacal, but I think these agents can stop you from getting actual work done. When I tried using these agents for my work, I ended up solving a lot of problems, but none of these problems seemed very important in retrospect.

Clearly that is a skill issue. I have no doubt that I’ll get better at it. And if your work is mediated through screens and you’re good at defining what you do and don’t like, these agents may be great for me.

But at the same time, it feels like a general manifestation of any sort of “life-improving” technology, which is often just about channeling of mental disturbances. So, no, I am not banking on it making me a Nietzschean ubermensch next month; nor helping me start a billion dollar company, or even on having a better time. Right now it still feels net zero: for every bit of busy work that it may rescue me from, it feels like it has potential to rob meaningful work or meaning---or maybe even my life of life itself. “Projects” that I really care about in my life are not app-shaped or list-shaped, and in doing things, technology is always an afterthought.

Rebecca Lowe on what you learn reading Wittgenstein’s notebooks.

From Transformer, an AI newsletter by Shakeel Hassim: When and if Anthropic IPOs, effective altruism causes are likely to be absolutely flooded with cash. It’s a good time to be a trustworthy advisor to friends who are smarter and more successful than you…

I don’t read the news unless I absolutely have to, but if I did, I would follow The Update by Stefan Schubert. Here he is briefly summarising the debate about how fast AI is progressing, and the relevance of this to the METR time horizon graph.

The Astera Institute is running an essay competition for researchers to write about the bottlenecks to science. Apply by May 1st, the top prize is $30,000.

From Barra Roantree: In the absence of a proper land tax, it’s a no-brainer that Ireland should be making relatively greater use of property taxes. Instead, it’s been continually eroded and degraded.

Stian Westlake responds to Sam Enright on citizens’ assemblies.

A reading list to understand whether large language models are engaged in introspection.

Gavin Leech’s greatest albums of 2025. There are also deaths of 2025, and the best blogs and essays of 2025.

What Rob Long read and liked in February. It seems that the story of the Buddha reached us in medieval Europe?

Barlaam and Josaphat: a popular Christian tale about an Indian king who leaves the palace to seek spiritual truth. Sound familiar? Beloved in medieval Europe, historians have since worked out that it descended from the story of the Buddha: Buddha/Bodhisattva → Middle Persian Bōdāsaf/Būdāsaf → Arabic Yūdhasaf/Būdhasaf → Georgian Iodasaph → Greek Ioasaph → Latin Josaphat.

Tom McCarthy on the lies we tell ourselves on useless days. I have had several of these recently.

Apply in 10 minutes for the Eigenprize, a $100,000 grant to do ???. Applications close on March 31st.

I titled one of the sections in my Notes on Taiwan ‘Cooking your own food is a policy failure’. One of the explanations I failed to consider was that we have been systematically underestimating the marginal cost of public funds, and that taxes distort the economy toward household production (in this case, cooking at home instead of paying someone).

What the Institute for Progress has been up to.

Congratulations to the new cohort of Emergent Ventures winners.

Tom Cunningham on the economics of transformative AI.

Neil Scott on the anthropology of Bradford.

How Francis Bacon read books. I have a Bacon quote on my (horribly outdated, in need of improvement) bookshelf page on my personal website.

The origins of Chinese versus European dragon myths, the strange quality of Ezra Pound’s translation of the Tang poet Li Bai, the mendaciousness of The Guardian, the first word of Beowulf, ICE agents storming consulates, Irish butter vouchers, and the theology of Quakerism. What could this be, if not what Peter McLaughlin read and thought about in January? We also have Peter’s thoughts prompted by having a baby:

My hypothesis is that, in a large proportion of cases, risk-aversion from pregnant women is neither a biological imperative nor a deep-seated internalisation of patriarchal attitudes, but rather is a sensible and adaptive tactic in an extremely low-quality (even adversarial) information environment.

I greatly look forward to having a child, so that I too can publish a victory lap about all the debates in the history of philosophy and linguistics in which this vindicates me. Peter is the only person I’ve known whose productivity went up immediately after having a baby. Here he is also (mildly) contra Emily Oster as a sensible source of parenting recommendations.

Podcasts

What Henry Oliver learned by reading Peter Pan to his children. I often wonder about how much further I might have gotten in life if I had a voice and accent as marvellous as Henry’s.

Finn Murphy and Tom McCarthy interview Scott Aaronson about quantum computing. Finn and Tom work at Nebular, a cool VC fund in New York (and now Irish Maths Olympiad sponsor!).

Adam Brown on general relativity, hitchhiking, and whether the nuclear bomb that fell on Nagasaki was dropped against direct orders. I find it hard to reconcile the way Brown discusses AdS/CFT correspondence as an occasionally useful toy model with Julian Gough’s hostility toward it.

A conversation between Dwarkesh Patel, Sholto Douglas and Trenton Bricken about whether reinforcement learning and LLMs are enough to get us to AGI. Has anybody written up a proper model in which, if fine motor skills are the last activity to be automated, the last jobs humans will have is assembling finicky parts on conveyor belts in windowless factories?!

Helen Castor on Richard II, Henry IV, and the political economy of medieval England. A very good discussion.

From the Sinica Podcast: Afra Wang explains The Morning Star of Lingao (临高启明), the influential collectively-written Chinese webfiction.

David Commins talks with Tyler Cowen about Saudi Arabia, Wahhabism, and the Gulf States. A good pairing with Matt Lakeman’s Notes on Saudi Arabia. Alas, I still have never been to the Middle East.

Music

Some new albums discovered, or listened to properly, this month:

Herbie Hancock, Thrust. Actual Proof is my favourite track here. In some ways, this is one of the less interesting Herbie albums, in that it is largely a continuation of Head Hunters, without the absurdly iconic bangers, but less avant-garde than the Mwandishi period.

Keith Jarrett, Standards in Norway. I did not recognise many of these melodies, except for All of You by Cole Porter (my favourite track here). Most are from the Great American Songbook era. I need to get around to finishing Ted Gioia’s book of jazz standards…

David Bowie, Blackstar. This was the album released two days before Bowie died in 2016, a collaboration with some very serious New York jazz musicians. The title track is long, dissonant, and quite complex. This is very good stuff. I suppose the lesson here is “The greats are always greater than you think”...?

Sister Irene O’Connor, Fire of God’s Love. Given how much of the great music traditions of the past were religiously inspired, it seems a shame that religious music has diminished so much in status and prominence. The artist is Sister Irene O’Connor, an Australian nun (!) who sings and plays all the instruments. The grooviest track is the adaptation of Luke 12:49, which (per Wikipedia) contains elements of dub. The Emmanuel Mass is also compelling, even if the 1973 drum machine is dated. There seems to be almost no record on the internet that it even exists, but every person educated in an Irish national school knows the Catholic music of the Alive-O programme, which this reminds me a bit of.

Now that I’ve reviewed so many albums in these posts, Claude has gotten extremely good at answering prompts about what I should listen to next, and where to fill in holes in my musical knowledge. The returns to having been a well-regarded (I hope!) blogger in the early 2020s are absurdly high.

Papers

Various, Charting Ireland’s Research and Innovation Future. This is the strategy document that was just released by Research Ireland, the main Irish science-funding body, which outlines their philosophy for the next five years. I think it’s pretty good as these things go: there is a commitment to start a metascience unit,¹ and the appendix also discusses how they are inspired by ARPAs such as ARIA and SPRIND.

Naturally enough, the devil is in the details, and I may write something more substantive about this for Progress Ireland. On page 33, there is a funny typo: Like everyone else, Research Ireland had been incorrectly assuming that ‘Nvidia’ was an acronym, and should thus be capitalised.

On page 34, there is a paragraph about existential risks, which (unsurprisingly) does not even mention risks from superintelligent AI as a candidate. Sad!

John McCarthy, Programs with Common Sense. I am nearing the day that I can truthfully tell people at Bay Area house parties that I am into AI, if by “AI” they mean symbolic AI and other approaches that are actually completely useless and only of historical interest.² As I understand it, McCarthy represented a strand within the style of thinking at the Dartmouth conference in 1956 that emphasised formal logic, more so than, say, the heuristic approach that Newell and Simon took with the logic theorist. He proposes a system here called the Advice Taker, an ancestor of automated reasoning. This version of the paper has a discussion session at the end, which begins with a funny referee report:

Dr McCarthy’s paper belongs in the Journal of Half-Baked Ideas, the creation of which was recently proposed by Dr. I. J. Good.

Ray Solomonoff, An Inductive Inference Machine. As a teenager, I first came across Ray Solomonoff in the same way as many people do: By seeing people on LessWrong completely overapply the concept of “Solomonoff induction”.³ I still don’t understand what that is. I spent some time over the weekend hitting my head against the Solomonoff convergence theorem, which says that the expected cumulative Kullback-Leibler divergence⁴ between the predictions made by the universal prior and the true environment is bounded by the Kolmogorov complexity of the true environment, up to a constant. That provides the formal justification of thinking about Solomonoff induction as an “ideal reasoning” process. I wrote a small bit about this in my bachelor’s thesis, in the context of whether the solution to certain paradoxes of expected value given the possibility of infinities might involve downweighting such bizarre outcomes for being highly computationally “expensive”.

All of which is a preamble to saying that I didn’t really understand this paper either. Interestingly, it’s to a significant extent a response to Noam Chomsky’s work that year (1957) on generative grammar.⁵ The only other thing I understood was catching a reference to Russell’s paradox,⁶ which made me smile.

Milton Friedman, Playboy profile. Say what you want about Playboy, but could you even imagine a porn mag in the year of our lord 2026 publishing a ten-page detailed and balanced interview with an economics professor? The Playboy profile was conducted at Friedman’s second home in Vermont over three days. It’s also interspersed, naturally enough, with adverts for the Air Force and cigarettes.

This interview is a good example of my allegation that “Friedman just isn’t nearly as interested in the details of how institutions work as I want him to be.” He says: “Government is failing at a lot of these things that it ought to be doing because it’s involved in so many things it shouldn’t be doing.” But like… is this really the way it works? Are we sure that, if we restrict the number of areas in which the government is active, its quality will necessarily improve in the remaining areas? Or will it be like DOGE, where attempted shrinkages will have catastrophic effects with minimal savings? In everything I read of Friedman –which was far from comprehensive, but was definitely a lot – I saw zero attempt to make this kind of idea rigorous, or write down a formal model. In the immortal words of anonymous Tumblr commentator:

people b saying things so definitively. like man i think it depends

In the interview, Friedman advocates for deregulating the airlines, drawing on the famous example of how flights within California were much cheaper than equal-distance flights between states, which were regulated by the Civil Aeronautics Board.

Famously, the top marginal rate of income taxation used to be very high in the US and UK after WW2 (peaking at 94% in the US, 95% in the UK).⁷ Rich people found lots of loopholes around actually paying this. Friedman says that the income tax schedule was not far off being flat in practice; this seems to be one of the reasons why he favoured abandoning the special exemptions and switching to a uniform rate of income tax. Frankly, I’ve run out of energy to fact-check Friedman’s exact claims against the Piketty/Saez literature about the de facto progressivity of American tax,⁸ but here is Opus 4.6 on how Friedman overstates his case.

As I pointed out in the main post, the early 1970s are an interesting period to look at for Friedman. By then, we had not seen the ‘liberal turn’; policy was still going in the opposite direction to what he would have hoped. Having initially signalled some allegiance with Friedmanite ideas, Richard Nixon yielded to overwhelming pressure and implemented a disastrous set of wage and price controls to tame inflation in 1971. He also imposed a 10% tariff on all imports to the United States (sound familiar?). But in Playboy, Friedman defends Nixon for acting properly, given the strength of feeling from the public.

(What?! Milton Friedman said that sometimes you need to compromise with public opinion and implement protectionism and price controls? Milton Friedman??)

Friedman, of course, strongly opposed the minimum wage. And the argument against it is compelling: benefits for workers are modest at best, and it creates unemployment. It is logically possible for a minimum wage to be net beneficial, if labour demand is relatively inelastic, or if employers are a monopsony that would otherwise be producing inefficiently little output. Friedman says that we know these countervailing effects don’t matter because of the following anecdote: In 1956, the unemployment rate for black and white teenage boys was the same: 8% (compared to overall unemployment of 4%). And then minimum wage was increased, and the unemployment rate among white teenage males rose 5 percentage points and over 14 percentage points for black male teenagers.⁹ He claims that the minimum wage is the primary explanatory factor.

On this topic, my vibe currently aligns with Nicholas Decker (which is not something I often say). Of all the economic policies in the world, the minimum wage is not among the most important. The empirical literature about it is so ambiguous primarily because, well, different labour markets are different.

But seriously, thinking that the minimum wage is important enough, one way or the other, in order to be responsible for white young unemployment increasing by a half and black young unemployment almost tripling is just crazy.

Clemens Fuest, Andreas Peichl, Sebastian Siegloch, Do Higher Corporate Taxes Reduce Wages? Micro Evidence from Germany. A nice paper; see my Progress Ireland post for an explainer.

Trevor Chow, Basil Halperin, Zach Mazlish, Transformative AI, Existential Risk, and Real Interest Rates. I finally got around to reading this paper, which has been on my list for so long that it has gone through multiple iterations in the interim (I read the version from October 2025). Here are the LessWrong comments and EA Forum comments. There are also arguments against by Jakob Graabak.

The basic idea here is that, if you expect to consume much more in the future and/or be turned into a paperclip, then you should borrow and consume more now. Thus, the real interest rate should rise sharply. We are not seeing this. The market does not seem to be expecting transformative AI even within thirty years. So, to the degree that you believe the efficient market hypothesis, you should believe we aren’t getting AGI anytime soon.

Nicholas Decker says that this line of thinking doesn’t really make sense, because AI will primarily affect us through creating goods and services that are not meaningfully available today at any price. It’s possible that $1 will provide so much more utility in the post-AGI world that an efficient market would be producing a savings glut and falling real rates now. It’s also infamously difficult to reason backwards from real rates; here is Basil on why he thinks his paper doesn’t violate Cowen’s Third Law (“all propositions about real interest rates are wrong”).¹⁰

At the same time, it feels like we should be able to say something useful about interest rates, because economics tends to understand risk and consumption smoothing pretty well. If I understood correctly, the effect that AI will have on either GDP or equity prices is much more unclear.

I have no strong opinions about any of this, but it seems like an important and fruitful line of research. My EA friends who think that this paper is just a lazy way for sceptics to entirely dismiss AI x-risk concerns are very, very wrong. As always, I am excited to see what these chaps come up with next.

Ashish Vaswani et al., Attention is All You Need. I finally read a foundational document of the modern age. Definitely read The Illustrated Transformer and watch the 3Blue1Brown deep learning series before you even attempt this. I’m not sure it really makes sense to be reading papers like this, as opposed to either older foundational textbooks, or more recent research papers. But I’m also not sure what one should be reading (is anyone?). The attention mechanism is simpler than I was expecting. The transformer attention formula is beautiful, and I enjoyed stepping through it.

This was a good example of what has been, so far, my #1 use case for Claude’s memory feature: to create multiple-choice exams for me about technical topics we’ve discussed in recent chats. I find that the multiple-choice format, in which I am forced to answer, is absolutely essential, versus fooling myself into thinking that I could have produced a good answer to an open-ended question.

Alec Radford et al, Improving Language Understanding by Generative Pre-Training. This is the GPT-1 paper from 2018, which was the next major milestone after Vaswani et al. showed that you could get shockingly good performance on an autoregressive model with only the self-attention mechanism and not bother with convolutions or recursion. This paper was the proof-of-concept that you could fine-tune on specific tasks, after a transformer was pre-trained on unlabelled text. I’ll have more to say about it soon.

Books

Jimmy Soni, Rob Goodman, A Mind at Play: How Claude Shannon Invented the Information Age. This book seems to be an instance of the “string theory is like a taco” model of science communication. It contains sentences like the following (page 64):

Alan Turing published a famously critical step toward machine intelligence. He had proven that any solvable mathematical problem could, in principle, be solved by machine.

One assumes the authors must be referring to Turing’s computable numbers paper. It’s unclear what “solvable mathematical problem” is supposed to mean, which is not what Turing’s paper is about. The paper is generally seen as a strong piece of evidence in favour of the Church-Turing thesis, which says that anything effectively computable can be computed on a Turing machine. But Church-Turing is, by definition, an unprovable philosophical position, and not something that Turing claimed to be able to “prove”.¹¹ The authors’ claim is especially strange, given that a later reformulation of the paper showed that the halting problem cannot be solved by a “machine”, even though it is (in some sense) a “solvable mathematical problem”.

It feels mean-spirited to continue to poke other holes in the book like this, because I appreciate the authors for trying to learn a new area and write the book they wish existed. But what resulted is really more about Claude Shannon’s vibes than anything else. This book is strongest talking about Shannon’s many interesting side projects, like the Roman numeral calculator (THROBAC) and robot mouse that learns how to navigate a maze (Theseus).

Milton Friedman, Rose Friedman, Free to Choose: A Personal History. All my serious thoughts are in the main essay, but here are the remaining tidbits I have in my notes:

1. Friedman is a writer who has genuinely expanded my vocabulary; who still says eleemosynary?

2. American revolutionary monetary policy was highly inflationary (page 347):

[T]he phrase “not worth a Continental” is a reminder of how that fiction [that money has value] was destroyed for the Continental currency issued in excessive amount by the U.S. Continental Congress to finance the American Revolution.

3. I didn’t know that tobacco was used as the currency of Virginia from the early 17th to mid-18th century, which led to inflation due to overproduction (since you could literally just grow more money). This nicely illustrates the problems with commodity money and other monetary principles like Gresham’s law.

4. Arguably one of the worst instances of regulatory capture in American history is the Interstate Commerce Commission, which was exploited by special interests to limit competition in the railroad industry. With time, its remit grew to include other forms of transportation such as trucking. The TV show version of Free to Choose interviews truckers who service busy routes, who were letting their trucks go idle because they couldn’t get an ICC license. Per Wikipedia, the ICC responded to competition from the monorail by… claiming its inventor was insane??

In March 1920, the ICC had Eben Moody Boynton, the inventor of the Boynton Bicycle Railroad, committed as a lunatic to an institution in Washington, D.C. Boynton’s monorail electric light rail system, it was reported, had the potential to revolutionize transportation, superseding then-current train travel. ICC officials said that they had Boynton committed because he was “worrying them to death” in his promotion of the bicycle railroad. Based on his own testimony and that of a Massachusetts congressman, Boynton won release on May 28, 1920, overcoming testimony of the ICC’s chief clerk that Boynton was virtually a daily visitor at ICC offices, seeking Commission adoption of his proposal to revolutionize the railroad industry.

5. The strange history of greenbacks, the first paper money issued by the federal government, introduced to finance the Union government in the Civil War (page 424):

During the Civil War, Congress authorized greenbacks and made them a legal tender for all debts public and private. After the Civil War, in the first of the famous greenback cases, the Supreme Court declared the issuance of greenbacks unconstitutional. One “fascinating aspect of this decision is that it was delivered by Chief Justice Salmon P. Chase, who had been Secretary of the Treasury when the first greenbacks were issued. Not only did he not disqualify himself, but in his capacity as Chief Justice convicted himself of having been responsible for an unconstitutional action in his capacity as Secretary of the Treasury.” Subsequently an enlarged and reconstituted Court reversed the first decision by a majority of five to four, affirming that making greenbacks a legal tender was constitutional, with Chief Justice Chase as one of the dissenting justices.

6. Free to Choose has an appendix in which he includes the American Socialist Party platform of 1928, showing that by 1980, America had either fully implemented or was on its way to implementing most of it (e.g. unemployment insurance). Dare I say: One man’s modus ponens is another man’s modus tollens?

7. Due to a lack of time, I had to cut a discussion from my session about how the Federal Reserve structure has weird borders that only imperfectly line up with state boundaries. That means that there are some states (e.g. Mississippi) divided into different regions whose only policy differences are monetary. That produced an amazing natural experiment, which provides evidence for the Friedman-Schwartz theory of the Great Depression that they outline in The Monetary History of the United States.

8. The first major failure in the series of bank runs that Friedman and Schwartz argue caused the Great Depression was that of the Bank of the United States. One reason why it was particularly damaging is that, because of its highly generic name, people assumed that it was a public bank, even though it was privately owned. Also, it was Jewish-owned; some historians have argued that anti-semitism is part of the reason why the bank was allowed to fail.

9. In Free to Choose, chapter six (on education), there is much discussion about the chasm in quality between good and bad public schools. The best American public high schools are extremely good: they are able to price out weak or troublesome students through a dystopian school districting system. One can make the case that the schooling system is already private, with the tuition paid implicitly through higher land prices. If you ask the average social democrat about this, I expect they will say this is because wealthy areas pay a lot in property tax and other local taxes, that mean that their schools are well-resourced. This is a common misconception: the American public school system is slightly progressive. Bad schools somehow cost more money to run than good schools. Several American states spend more on public education than any European country, and get abysmal outcomes. There is obviously a lot to be said about this topic, but I find it difficult to believe that the problems of education have much at all to do with insufficient resourcing for the public system.

10. An eyebrow-raising comparison Friedman makes about education is where he contrasts the graduation rate at Dartmouth (a private university) at 95% to the graduation rate at UCLA (public) at 50%.¹² These numbers are extremely outdated: UCLA is now an extremely selective institution at which over 90% of undergraduates finish. But using the numbers as they were at the time, Friedman is trying to make a point about how private institutions are better incentivised to make sure that students finish their education than public ones. My first instinct is that selection bias is the most powerful force in education, and that graduation rates are primarily determined by which kinds of students choose to apply to different universities in the first place. The papers I can find about this conclude that, after adjusting for student quality, public universities are actually better than private ones at getting students to graduate.

11. Notable in its absence from Friedman’s thought is much discussion of colonialism. Episode one of Free to Choose speaks effusively about British-owned Hong Kong (the polls, insofar as they exist, indicate that over 90% of Hong Kongers would vote to be re-colonised by the British). He also wrote effusively about the free trade policies of Meiji Japan, without mentioning that they were imposed by threat of invasion from the U.S. and U.K. Friedman obviously knew about that, but it’s a strange choice of emphasis.

12. From chapter three, the highly authoritarian nature of the end of the gold standard in the US does not cease to amaze: In 1933, private ownership of gold was made illegal, and you had to surrender your gold to the Federal Reserve for a fixed (low) rate. The treasury revalued it at a much higher rate a year later, and kept the difference. If a Latin American country did something like this, what would we call it? What would we call its leader?

13. In reading up more about school vouchers, I learned that there was previously a Friedman-style voucher system in the United States: in the South, to avoid racial integration.

Films

Arthur Miller, The Crucible (play). Yes, that is the Arthur Miller, patron saint of socially awkward nerds with considerably more attractive wives. Here he is on why he wrote The Crucible. A good play and production, which I went into with no context. I wonder what fraction of the Irish audience didn’t even catch that this is about Joseph McCarthy and HUAC? The theatre pages have tended to be negative on this production for not being heavy-handedly political enough. Alas.

Bong Joon-ho, Incoherence. Bong Joon Ho’s graduation project from the Korean Academy of Film Arts. It’s half an hour long, consisting of three vignettes and an epilogue. The first is good, the second and third drag, and the epilogue is fantastic. You can certainly see early manifestations of Bong’s somewhat reactionary obsession with class, but overall, my conclusion is that Bong is someone who iterated and got better through practice, rather than a natural genius.

Óliver Laxe, Sirāt. Here is the review from Tyler Cowen, who calls it one of the five or six best films of the millennium. Here is a spoiler-filled post on what it really means. This is among the two or three films I’ve been most moved by in the last year. The big screen is essential.

Various, Six Nations: Ireland v Scotland. Even for someone who doesn’t really understand rugby, this was very good. It resulted in Ireland winning the Triple Crown, the trophy that goes to any Home Nation if they beat all of the others. Alas, Ireland lost the tournament to France.

Milton Friedman, Free to Choose. You can watch this entire ten-part series for free on YouTube. It’s genuinely well-done, and I wish I had watched it before my event. In episode four, the person on welfare in Britain interviewed about their long-term dependence on it has a thick Irish accent, which, uhm, probably didn’t do any favours for integration.

There are some very fun filming locations here. In episode eight, Friedman goes to the island of Kos to talk about Hippocrates, and how his famous oath is flagrantly violated by medical licensing. There are several “that was then, this is now” moments: in episode seven, Friedman says he only sometimes wears his seatbelt.

It’s funny how many people I recognise unexpectedly pop up in these episodes. Peter Temin is featured in episode three (his hair was already grey in 1980).

From YouTube, we have De La Soul, Kamasi Washington, Ezra Collective, and Snarky Puppy on NPR Tiny Desk. There is also Dwarkesh’s statement about why the Department of War fiasco is so damaging. Finally, we have this informative video about why Dublin needs a metro system. This may be good to combine with Brian Potter’s Works in Progress essay about why metro systems switched away from ‘cut and cover’ and toward tunnel boring machines. The people yearn for economical tunnel construction methods.

1. Or rather, a “meta-research unit” (page 51).

2. To clarify: I have no idea whether symbolic AI is, in fact, useless.

3. Solomonoff induction assumes that the data-generating environment (“The Universe”) is computable. I have no strong opinion about whether the universe is computable.

4. I first learned about KL-divergence from a paper called Information Theory for Intelligent People, but it’s proved surprisingly helpful.

5. This predates the ‘universal grammar’ idea of the 1960s. Syntactic Structures, Chomsky’s 1957 book, introduced the famous “colourless green ideas sleep furiously” example. We have a fun idea for a Noam Chomsky reading group, in which the first half is strictly linguistics, and then we’re only allowed to talk about his politics in the pub after once everyone is smashed.

6. Page 5: “One of the limitations of the transformation and combination methods discussed thus far, is their inability to deal with higher order sets - sets whose members are themselves sets. Such devices are of much importance in mathematics, in logic, and in science. There are some logical difficulties that may arise if sets are allowed to be members of themselves.”

7. For context, the highest marginal rate of income tax proposed by Bernie Sanders was 52%.

8. Or indeed, to fact-check the Piketty/Saez literature against, uhm, reality. I had lunch today with a very sharp public finance economist, who said that Thomas Piketty has the same reputation in his profession as Friedman, as a genius crank.

9. In my favourite quote about this, he says “If we’d had minimum-wage laws and all the other trappings of the welfare state in the 19th century, half the readers of Playboy would either not exist at all or be citizens of Poland, Hungary, or some other country. And there would be no Playboy for them to read.”

10. They have two ways of measuring real interest rates: taking nominal rates and subtracting a measure of expected inflation, and also looking at the yields on long-term inflation-linked bonds. Encouragingly, these give the same answer.

11. Turing himself did not believe that the human brain is limited by that which is Turing computable.

12. Free to Choose, page 251.