Sam Enright is engaged in innovation policy at Progress Ireland, an independent policy think tank located in Dublin. He also oversees a publication named The Fitzwilliam. Most relevant to our discussion, on his personal blog, he publishes a well-received link roundup. Below is a condensed version of his Links for June.
Blogs and Short Links
1. Henry Oliver asserts that the notion of studying English literature brilliantly equipping someone to become a prime minister is absurd.
2. Patrick Collison provides insights on Hibernofuturism.
3. José Luis Ricón discusses Claude’s constitution, also linking to his additional resources. Interestingly, he delves into debates on whether worms can acquire memories through consuming other worms.
4. Although Goodreads has numerous documented issues, at a conceptual level, it perhaps suffers the most due to its general recommendations which presuppose that the “average reader” is a valid benchmark. Personally, I find it irrelevant that Why We Sleep by Matthew Walker ranks in the top 5% of ratings. To counter this, Mark Greville has developed BookTribes, which recommends books based on user taste similarities rather than popularity.
5. Congratulations to the launch of the Arq Foundation, a new think tank in Brussels dedicated to preparing Europe for transformative AI.
6. The latest on “unexpected Timothy Williamson” caught my attention. Someone should send Ava a copy of the Timothy Williamson crochet puppet.
7. I often get asked if there’s a definitive book on the Irish housing crisis that isn’t riddled with economic inaccuracies. Currently, there seems to be a gap. However, I expect Ronan Lyons will address this with Brittle: How the Housing Crisis is Breaking Ireland—and How to Fix It, now available for pre-order, set for release in October.
8. Alex Tabarrok has refuted the Coase conjecture.
9. The literary genius Jorge Luis Borges would have appreciated experimenting with large language models, as many of the motifs in his works are anticipated in his fiction. For Borges enthusiasts, Pierre Menard, Author of the Quixote is a notable reference. I couldn’t help but laugh while reading Linch Zhang’s efforts to fine-tune an LLM to replicate the original story token-for-token.
10. Ben Yeoh publishes a weekly letter that I find particularly insightful. His latest update on whether cancer outcomes are genuinely improving (after accounting for relevant statistical adjustments) shows a decrease in cancer mortality by around one-third since the early 1990s.
11. A new magazine, In Development, recently launched, featuring an essay by Paul Niehaus discussing the origins of GiveDirectly. I found footnote 13 particularly witty, echoing my own skepticism about microfinance:
In 2005, my co-founders and I managed to secure invitations to a kickoff event for the International Year of Microcredit at the United Nations. The cocktails were, if I remember correctly, stronger than the evidence.
In Development’s editor-in-chief, Lauren Gilbert, also curates a linkpost, directing me to Chris Blattman’s list of books that development economists ought to explore.
12. In my essay about Milton Friedman, I remarked on the lack of skepticism regarding a well-known manipulated image of him. A reader pointed out to me, via a Forbes journalist, that the image was indeed fake and Friedman found it amusing enough to share with his friends.
13. My all-time favorite linguistics book is The Language Instinct by Steven Pinker. Here are some thoughts about it by Dwarkesh Patel.
Music and Podcasts
1. Dario Amodei recently appeared on the Dwarkesh Podcast. I felt compelled to comment, “that feeling when you’ve just wagered humanity’s future on a model resembling competing bottled-water companies from 1838”:
[L]et’s just abstract the whole industry here. Let’s just imagine we’re in an economics textbook. We have a small number of firms. Each can invest a limited amount. Each can allocate some fraction to R&D. They face some marginal cost to serve. The gross profit margins on that marginal cost are quite high due to efficient inference. There’s some competition, but the models are also differentiated.
Companies will compete to enhance their research budgets. However, because there are only a few players, we encounter what is known as the Cournot equilibrium. The point is it doesn’t settle into perfect competition with zero margins. Even with three firms operating rationally and independently, it doesn’t balance out to zero.
2. Tom Holland interviewed Paul McCartney. Each time I listen to McCartney, I can’t shake the memory of Simon Pegg cleverly explaining the embouchure in Beatles’ impressions.
3. In the ChinaTalk podcast, representatives from the Institute for Progress and the Foundation for American Innovation converse about the evolution of American think tanks.
4. Recently, I unearthed a fascinating podcast on Chinese history called Barbarians at the Gate. In one episode, they discuss the An Lushan Rebellion, a devastating civil war that marked the end of the Tang dynasty’s glory years.
Numerous internet posts note that historical lists of the deadliest events often highlight obscure Chinese civil wars.
For a while, I struggled to grasp the perplexing historical population estimates of China. When examining city sizes or casualty figures, they often seemed unusually out of sync with what one would expect from general technological advancement and other context.
As a teenager, I recalls being stunned by the death toll estimates of various Chinese civil wars while reading The Better Angels of Our Nature. In one passage, Steven Pinker notes that the An Lushan Rebellion “according to censuses, resulted in the loss of two-thirds of the empire’s population, a sixth of the world’s population at that time” (p.195).
Surely, that can’t be accurate, can it?!
These estimates stem from comparing population data between Tang’s 754/755 AD census (~53M) and the 764 census (~17M).
However, no reputable source believes that 36 million figure is credible as an accurate death toll. This seems more like a maximum estimate, reflecting dislocation and the administrative state collapse during that period. Historical data in the Chinese census dates back to 2 AD. Thus, equating the differences in census figures with deaths could mischaracterize more functional areas, given their capability to manage a census—a capability lacking in most regions throughout history.
I don’t believe anyone has thoroughly revisited Pinker’s examination of violence history with the skepticism deserved regarding census figures as upper bounds. Regrettably, many rebuttals to the Better Angels thesis appeared to be disorganized and incoherent.
Moreover, did you know that one of the most famous poems globally is about the An Lushan Rebellion? Most students in China can probably recite the opening lines of Spring View, where Du Fu reflects on the devastation caused by the rebellion:
The country is broken; though hills and rivers remain,
In the city in spring, grass and trees are lush.
This serves as quite a lengthy introduction to what, in the end, marked only my second favorite podcast episode on the An Lushan Rebellion I listened to last month. My top pick was Melvyn Bragg’s episode on In Our Time. When Melvyn mentioned Pinker and his reference to An Lushan in his work, it sparked unanimous laughter among the expert guests. Guffaws!
5. A discussion on the life and philanthropy of John Arnold captured my attention. Arnold began his career at the profitable (yet legitimate!) natural gas trading division of Enron. Following the collapse of Enron, he went on to establish one of the most lucrative hedge funds in history. At 38, Arnold returned the funds to his investors, focusing entirely on philanthropy.
6. Paul Gillingham shares insights on life in Mexico. Though I know little about Mexico and haven’t visited, I would happily spend my days walking around with my AirPods, absorbing podcasts like this to learn about various locations and their histories.
7. A new episode of 80,000 Hours discusses their rebranding to focus primarily on advising careers tied to the transition to AGI.
8. Gustav Mahler’s Symphony No. 5, along with associatedSticky Notes episodes, were enjoyed live under the baton of Jonas Alber. This formed the foundation of my first classical music discussion group, an experiment aimed at expanding the reading group format’s potential. You can check out my discussion prompts here. The third movement (Adagietto) is exquisite, but its connection with the others can be quite abstract, making it easy to get lost in Mahler’s works. Consequently, I conclude that this piece may be too complex to start with. I’ve listened to three of Mahler’s symphonies live, with seven more to go before I die.
9. One of my recent musical discoveries was Benjamin Britten’s Violin Concerto. Check out this YouTube version by Janine Jansen. Written in 1939, it’s often interpreted as an elegy for the Spanish Civil War. However, it strikes me as an expression of dread about an impending war. Britten isn’t typically among my favorites, but this piece harbors elements I adore, especially the pizzicato in the first movement.
As is my habit, I sought out Tyler Cowen’s perspective. In an article commemorating what would have been his 100th birthday, he discussed how Britten is essential for grasping John Maynard Keynes’s economic thought and “the broader history of homosexuality in England” (surprise!).
Books and Papers
1. DeepSeek, DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning. The original DeepSeek paper released in January 2025 had a significant impact, triggering a downturn in American tech stocks as a Chinese language model demonstrated near frontier performance while being dramatically more efficient computationally. It’s astonishing that this occurred only a year and a half ago; it seems like an eternity.
DeepSeek-R1 was the second ‘reasoning’ model following OpenAI’s o1. Before its introduction, improved language models had not effectively increased the quality of responses per token. With reasoning models, reinforcement learning allows for upweighting the tokens in the “chain of thought” leading to accurate answers. This explains why LLMs began stating things like “wait, let me reconsider…” and correcting their previous outputs.
My discussion prompts for the Fitzwilliam AI Circle regarding the DeepSeek paper can be found here.
2. Amanda Askell, Claude’s Constitution. I’m likely late to the conversation, given that this document generated buzz in January. It remains challenging to assess the significance of this constitution, particularly as the specifics of its integration into the training process remain hidden. When Claude makes moral judgments, how much is an emergent effect of pretraining, contingent on the system prompt, influenced by human feedback reinforcement, or due to Anthropic’s unique approach to “model character”? Is there clarity on this for anyone outside the labs? What about within them?
From what I gather, the constitution aims to establish a core set of stringent constraints incorporated into the training process, ensuring the model exercises sound judgment regarding which types of user requests to decline or challenge, while “understanding” the rationale for those judgments. This approach reflects Anthropic’s vision of humanity evolving alongside Claude. If we present numerous high-quality human reasoning examples, he might evolve to be wiser than us.
A hilariously absurd New Yorker essay about the Constitution came from an author who clearly didn’t bother grasping the technical aspects involved, aspects that are crucial. The argument was summed up as:
One way to interpret Claude’s Constitution is that it signifies the collapse of the state. The U.S. Constitution is evidently failing, leading to the necessity of Claude’s constitution, which purportedly safeguards American citizens (and foreign nations) from the overwhelming strength of the U.S. military.
What is this person talking about? Did the renowned New Yorker fact-checkers neglect to inform her that the Constitution was published only in January this year? Or that other labs release models (aside from Grok) that aren’t overtly less ethical without having a constitution?
Regardless, I noticed a Twitter comment questioning why copyright non-reproduction wasn’t included in the Constitution; when I queried Fable, it likened this to embedding a speed limit into a moral philosophy text. Copyright laws are historically contingent and jurisdiction-specific. While I’m not a champion of intellectual property anarchy, it seems imprudent to codify the legal framework of 2026 America into the eternal fabric of moral judgment—this indeed resembles placing a boot on a human’s face perpetually.
A light-hearted note: I happen to own a physical copy of Claude’s Constitution, where the essence of artificial intelligence is color-coded to #DE643C in hexadecimal.
3. Alan Ahearne, Entrepreneurial Activity and Living Standards in Ireland. This paper was authored by the former economic advisor to Micheál Martin, commissioned by John and Patrick Collison. Although I had no involvement in this work nor do I know Ahearne, I found its content quite relevant and decided to read it.
It’s important to note that, on average, Irish workers rank among the most productive globally. However, a closer examination reveals a stark contrast: an exceptionally productive multinational sector versus an unremarkable, potentially mediocre domestic economy. This disparity heightens the vulnerability of the Irish economy; a sudden departure of a few American firms could have catastrophic consequences. This scenario underscores the necessity to foster domestic entrepreneurship.
4. George Berkeley, The Querist. I sometimes wonder if I’m the only person on earth who hasn’t read any of Bishop Berkeley’s philosophical works while having delved into his pamphlets on Irish coinage from the 1730s.
Berkeley’s seminal philosophical work on idealism was crafted during his twenties. His life thereafter was intriguing, culminating in his role as the Bishop of Cloyne in County Cork, located in southwest Ireland.
The Querist stands out as a remarkable piece: a three-part pamphlet advocating for paper currency and a national bank within the context of Ireland’s development, all framed through sarcastic numbered questions. It’s akin to Wittgenstein’s Tractatus but with a focus on banking, credit, and occasionally agricultural productivity in 18th-century County Tipperary.
This book is so obscure that acquiring a copy proved challenging; however, I found the formatting of the Project Gutenberg edition entirely satisfactory. My comprehensive treatise on the Irish Enlightenment will materialize eventually, but for now, I assert that The Querist deserves recognition as one of the most overlooked works in the history of economic thought.
Films and Videos
1. Jonathan Demme’s Stop Making Sense is a legendary documentary showcasing a Talking Heads concert staged in a Hollywood theater back in 1983. They perform hits from across their album repertoire up to that point, including Burning Down the House and their rendition of Take Me to the River.
Endnotes
[1] Answer: probably not.
[2] One reason to be skeptical is the civil war fighting primarily occurred in the north. The only credible death estimates nearing two-thirds I’ve seen relate to Paraguay’s war with all its neighbors.
[3] This marks the third instance of Tang poetry featuring on my blog. It also prompted an engaging digression by Peter McLaughlin — one of my notable talents!
[4] Due to the controversies surrounding Enron, Arnold could only raise a limited amount of external capital.
[5] He recently recorded a podcast with Rebecca Lowe discussing Britten’s War Requiem, which I haven’t had the time to listen to yet.
[6] Earlier iterations of constitutional AI—in which models rank their responses against ethical principles—became standard after the publication of the Anthropic paper in late 2022. The general method introduced, Reinforcement Learning by AI Feedback, isn’t inherently linked to a written constitution or models’ ethical behavior and has simply been a common method of capability enhancement. I doubt this aligns with the author’s intent.
[7] For instance, two-thirds of the gross value added in the Irish economy originates from the multinational sector. Foreign-owned companies are six times as productive as domestically-owned counterparts, while a comparative group of countries shows a ratio of 1.8 times (page 16).
