Dwarkesh on slow AI take-off
I’ve probably spent over a hundred hours trying to build little LLM tools for my post production setup. And the experience of trying to get them to be useful has extended my timelines. I’ll try to get the LLMs to rewrite autogenerated transcripts for readability the way a human would. Or I’ll try to get them to identify clips from the transcript to tweet out. Sometimes I’ll try to get it to co-write an essay with me, passage by passage. These are simple, self contained, short horizon, language in-language out tasks – the kinds of assignments that should be dead center in the LLMs’ repertoire. And they’re 5/10 at them. Don’t get me wrong, that’s impressive.
But the fundamental problem is that LLMs don’t get better over time the way a human would. The lack of continual learning is a huge huge problem. The LLM baseline at many tasks might be higher than an average human’s. But there’s no way to give a model high level feedback. You’re stuck with the abilities you get out of the box. You can keep messing around with the system prompt. In practice this just doesn’t produce anything even close to the kind of learning and improvement that human employees experience.
The reason humans are so useful is not mainly their raw intelligence. It’s their ability to build up context, interrogate their own failures, and pick up small improvements and efficiencies as they practice a task.
Here is the whole essay, I am in agreement.
My Conversation with the excellent John Arnold
Here is the audio, video, and transcript. Here is part of the episode summary:
Tyler and John discuss his shift from trading to philanthropy and more, including the specific traits that separate great traders from good ones, the tradeoffs of following an “inch wide, mile deep” trading philosophy, why he attended Vanderbilt, the talent culture at Enron, the growth in solar, the problem with Mexico’s energy system, where Canada’s energy exports will go, the hurdles to next-gen nuclear, how to fix America’s tripartite energy grid, how we’ll power new data centers, what’s best about living in Houston, his approach to collecting art, why trading’s easier than philanthropy, how he’d fix tax the US tax code and primary system, and what Arnold Ventures is focusing on next.
Excerpt:
COWEN: Say there’s a major volcanic event, and there’s a lot of ash in the sky for two or three years. Solar needs a backup. In the meantime, before the volcanic event happens — and of course, that’s quite rare — how much do we need to be up and running with the backup energy infrastructure? What do we need for reserve capacity in case the solar goes down?
ARNOLD: Good question. It would be difficult. It’s doable today. I think as solar continues to grow in market share, both in the US and globally, it will have to be met with some type of battery, a significant battery resource. That’s part of the economics of solar now, that it’s not just sticking it right outside of Phoenix, but it is solar plus transmission or solar plus battery. The question of what happens in that type of event — it would be difficult. The existing energy infrastructure is still largely around.
COWEN: But it will dwindle over time, right?
ARNOLD: It will dwindle over time.
COWEN: Is there some market issue? Say the volcanic event is only once every 150 years, but sooner or later, one happens. In the meantime, you need economic incentives for the gas or the nuclear to be ready. Does our government just keep on paying for those for 149 years in a row until the catastrophe comes?
ARNOLD: It’s a great question, and I think this is why nuclear, and particularly next-gen nuclear, is considered the holy grail, right? You’re not constrained by location. You’re not constrained by, is the wind blowing, is the sun shining? And it’s a clean resource. The problem today is just economics. In order to develop the current generation of nuclear, it’s extraordinarily expensive. Next generation — either small modular fission or fusion — both have a number of technological as well as unclear economics in how they compete.
I do think this question of how do you do this transition in a manner that maintains affordability but continues to get cleaner and lower emissions over time is a complex one, and I think it’s one that the environmentalists probably oversold five years ago in saying that this was going to be an easy transition. It’s certainly not. Just the scale and scope of the energy system is enormous, as you’re pointing to in your question. The need for backup, the need for a diversity of fuels, and how they complement each other is real, and you can’t replace that just with the intermittent resources we have today, plus battery.
And:
COWEN: What’s your most optimistic scenario for the US energy future from an environmental point of view, something that could plausibly happen?
ARNOLD: I think next-gen nuclear, if we can overcome the technical hurdles, if we can overcome the economic hurdles.
COWEN: But isn’t NIMBYism the biggest hurdle? The others I could imagine overcoming pretty readily, but I live in Fairfax County, which builds a fair amount. People there just don’t want nuclear. It’s irrational, but I’m not sure they’ll change their minds. It could be called fusion; it’s still nuclear to them.
ARNOLD: Yes, I’ve been surprised. That was my prior five years ago. I’ve been surprised at the number of jurisdictions that are inviting these next-gen nuclear companies to come. Texas, for instance, just passed a bill creating new incentives for nuclear companies to come and build their first plants and pilot projects in Texas. You see jurisdictions that are choosing to take the economic growth associated with it and that have more of a building culture and say, “Come here.”
I think, as things get proven out, then the question is, will the Fairfax counties of the world see what’s going on and become more agreeable to having that? I think it’s very similar to self-driving cars.
There’re some jurisdictions that say, “Come here. We want you to come, test,” and this is what’s happening in Texas. These companies say, “We want you to come pilot your projects here.” And some jurisdictions are saying, “No, prove it out, and then we’ll talk.”
COWEN: My nightmare is that even Texas becomes NIMBY. You see this in Austin already. Houston, Dallas will become more like the rest of America over time, maybe even San Antonio someday, El Paso with more time.
Interesting throughout, recommended. We also talk about art and art collecting…
Using AI to explain the gender wage gap
Understanding differences in outcomes between social groups—such as wage gaps between men and women—remains a central challenge in social science. While researchers have long studied how observable factors contribute to these differences, traditional methods oversimplify complex variables like employment trajectories. Our work adapts recent advances in artificial intelligence—specifically, foundation models that can process rich, detailed histories—to better explain group differences. We develop mathematical theory and computational methods that allow these AI models to provide more accurate and less biased estimates of how much of group differences can be explained by observable factors. Applied to real-world data, our approach reveals that detailed histories explain more of the gender wage gap than previously understood using conventional methods.
That is from a new paper by Keyon Vafa, Susan Athey, and David M. Blei. Via the excellent Kevin Lewis. This is also real progress on the methodological front.
Wednesday assorted links
1. Australia may be passing a tax on unrealized capital gains.
3. The full Tic Tac 2010 report. Here is one summary. Here is a Grok summary. I will read it when back home.
4. Curtis Yarvin profile (New Yorker).
5. A new and different attempt to summarize Proust.
6. Jessica Crispin has reemerged with a book on Michael Douglas and the crisis of masculinity (NYT).
French fact of the day
De Gaulle was the target of about thirty serious assassination attempts, two of which — in September 1961 and August 1962 — nearly succeeded. For some anti-Gaullists, the fixation on de Gaulle became so incorporated into their personality that their original reasons for wanting to kill him were eclipsed by the hatred he inspired.
Hating de Gaulle for accepting Algerian independence was one of those motives for at least one of those attempts.
That bit is from Julian Jackson, A Certain Idea of France: The Life of Charles de Gaulle, a good book.
My days collecting Mexican art, part II
Recently I wrote about my quest to track down Mexican amate (bark paper) painter Juan Camilo Ayala, but there is another part to the early story, namely looking for his brother Marcial Camilo Ayala, also a painter.
Marcial no longer lived in Oapan, as he found village life intolerable. So he settled in Taxco (later Cuernavaca), and it was Juan Camilo who told me that when I showed up at his house in Oapan. Originally I was hoping to meet both brothers on that first trip.
When I arrived in Taxco on my next Mexico trip, I had the strategy of asking all tradionally-clothed women in the city center “do you know Marcial Camilo Ayala?” Far from being a needle in the haystack strategy, this yielded results within seconds. All of a sudden I was chatting with Marcial’s youngest daughter, Oliva. She in turn brought me down a steep cobblestone street to see Marcial, who was painting in a dark back room in Taxco. It all felt rather hopeless, at least at first.
Marcial and Juan were quite different. Marcial is by far the most intellectual person from Oapan, as he could speak at high levels about Picasso and Rousseau, Zapata and land reform, Nahuatl poetry, and the late string quartets of Beethoven (alas he passed away almost ten years ago). Juan cannot meaningfully read or write, but he is a corn farmer who knows everything about the rain. Marcial typically is considered the strongest painter from Oapan, and multiple times he had traveled abroad for exhibits of his work.
I now had two reasons to go to the region, namely Juan and Marcial. And so I became patrons of them both, and now have dozens of works from each of them, including some very large six foot by eight foot creations. I kept on returning to Guerrero, and would spend some time in Oapan with Juan and his family, and some time with Marcial, either in Taxco or Cuernavaca, typically talking about ideas and art. I finally started to learn proper Spanish from all the required back and forth.
In my time in Oapan I enjoyed the stars at night, the fiestas and processions, the long hours sitting around talking and joking with Juan’s family, and of course the food. The musty blue corn tortillas are to die for. If you want some fresh fish, great, but they have to go down to the river and catch it for you. The bean tamales and moles with pepitas are incredible. I once commissioned a barbecue meal, $80 for a full goat, cooked underground overnight, as from prehispanic barbeque traditions. Most meals did not involve meat, however, other than the staple of eggs.
Yet life in Oapan is not easy, not even for the visitor. There was no flush toilet or shower. The “bed” was a hard slab, and the evening temperatures inside the room exceeded one hundred degrees Fahrenheit. The roosters crow at 4 a.m., and then everyone is awake. You can leave, but within the Oapan of that time, dollars could not buy you conveniences. There is an ever-present risk of dengue and sometimes malaria as well.
I got to know the four main amate painting villages (Ameyaltepec, Xalitla, and Maxela are the others), and met virtually all the living amate painters of note. I visited the renowned Alfonso Lorenzo Santos, both chained to the wall in his home in Ameyaltepec and also in the mental hospital in Cuernavaca. (Alfonso was later profiled in The Wall Street Journal, and for that journalist, Bob Davis, I served as Mexico guide and translator.) Occasionally, when looking for new amates, I had to throw rocks at the wild dogs to make my way to the homes on the edge of town.
Over the course of about a dozen years of visits, I built up what is the world’s largest and I would say best amate collection, with hundreds of quite distinct works. I also managed to buy an important early private collection, from the 1980s, with more than two hundred paintings. For years I tracked all the amate painting listings on eBay, snagging many a bargain. Later I served as (unpaid) amate painting consultant to the Smithsonian, when they set up the American Indian museum now on the mall. I am pleased that the assemblage of these works is preserving a significant cultural episode and tradition in Mexican history.
I also collected a good deal of village ceramics, still done with red clay using pre-conquest methods, noting that not all of them made it home intact. The Spanish word “burbuja” — bubble wrap — remains prominent in my mind and vocabulary. Ideally, I would like to do a major “air lift” of traditional pottery out of Oapan, but these days the drug gangs are a major obstacle.
Buying art works from Juan and Marcial also evolved into charity, and I developed my thoughts on direct cash transfers. I wrote those up on MR long ago, and I am pleased to report they had some influence in inspiring the non-profit Give Directly.
Eventually I wrote a whole book on the economy and polity of Oapan, and on the lives of the amate painters. It was published with the University of Michigan Press under the title Markets and Cultural Voices: Liberty vs. Power in the Lives of the Mexican Amate Painters. It has sold the least well of any of my books, by far, but it is one of my favorites and it is quite unlike all the others.
Over the years, there was one amate painter whose works I never tracked down, namely Jesus Corpos Aliberto. Marcial had told me he heard a rumor that Jesus Corpos was living in a dumpy hotel in the middle of Mexico City, Hotel Buenos Aires. I found my way to the hotel, and yes Jesus was there with a big stack of brilliant amates he was looking to sell. They let him stay there in a smelly back room. Sadly he was insane, and would sell the amates only for millions of pesos. During yet my next trip to Mexico City, I returned but the hotel was gone altogether, eliminated by gentrification. I had no remaining links to Corpos. At that point, and following the passing of Marcial, and the aging of the other main amate painters, that part of my life largely was over. And so my story with amate painting ends with the same basic obstacle it started with: a stubborn refusing to sell me something, thwarted markets in everything.
China markets in everything
But in the country’s large cities, spaces that offer the solution have begun to spring up: companies that allow people to pretend to work.
For a daily fee of between 30 and 50 yuan ($4-$7), these companies offer desks, Wi-Fi, coffee, lunch, and an atmosphere that mimics any work environment. According to a report in Beijing Youth Daily, although there are no contracts or bosses, some firms simulate them: fictitious tasks are assigned and supervisory rounds are even organized. For a fee, the theatricality can reach unimaginable levels, from pretending to be a manager with his own office to staging episodes of rebellion against a superior.
Zonghua is Cantonese and prefers not to give her real name. Tired of traveling and the pressures of the financial world, she resigned from her position in the spring of 2024, she tells this newspaper via a local social media platform. “I was looking for a more stable life,” she writes. But she doesn’t dare tell her family the truth. At first, she went to libraries, but for the past few months, she has been paying a monthly fee of 400 yuan ($55) for a comfortable space to spend the day; it’s much cheaper than spending hours in a cafe. Zonghua doesn’t know how much longer this situation will last, as, for now, she’s not having any “success” with her applications.
Here is the full story, not unrelated to UBI debates either. Via R.
Tuesday assorted links
1. The coalition against the Trump tariffs (NYT).
3. Alasdair Macintyre obituary (NYT).
5. “Conjecture: econ academia is very far from the optimal architecture for data research. The optimal architecture is much more centralized. Half the researchers work together, or in large teams, to build clean and efficient datasets. The rest work in smaller teams to analyze data” Link here.
6. In Burkina Faso, free contraception does not lower fertility much.
7. Good NYT piece on the new Torigian biography of Xi’s father.
Sentence of the Day
FT: Analysis by Torsten Sløk, chief economist at Apollo, suggested that US government credit default swap spreads — which reflect the cost of protecting a loan against default — are trading at levels similar to Greece and Italy.
Yikes!
Do more laws boost economic growth?
This paper analyzes the conditions under which more legislation contributes to economic growth. In the context of US states, we apply natural language processing tools to measure legislative flows for the years 1965–2012. We implement a novel shift-share design for text data, where the instrument for legislation is leave-one-out legal topic flows interacted with pretreatment legal topic shares. We find that at the margin, higher legislative output causes more economic growth. Consistent with more complete laws reducing ex post holdup, we find that the effect is driven by the use of contingent clauses, is largest in sectors with high relationship-specific investments, and is increasing with local economic uncertainty.
That is from a new issue of the JPE, by Elliott Ash, Massimo Morelli, and Matia Vannoni.
What Explains Growing Gender and Racial Education Gaps?
In the 1960 cohort, American men and women graduated from college at similar rates, and this was true for Whites, Blacks and Hispanics. But in more recent cohorts, women graduate at much higher rates than men. Gaps between race/ethnic groups have also widened. To understand these patterns, we develop a model of individual and family decision-making where education, labor supply, marriage and fertility are all endogenous. Assuming stable preferences, our model explains changes in education for the ‘60-‘80 cohorts based on three exogenous factors: family background, labor market and marriage market constraints. We find changes in parental background account for 1/4 of the growth in women’s college graduation from the ’60 to ’80 cohort. The marriage market accounts for 1/5 and the labor market explains the rest. Thus, parent education plays an important role in generating social mobility, enabling us to predict future evolution of college graduation rates due to this factor. We predict White women’s graduation rate will plateau, while that of Hispanic and Black women will grow rapidly. But the aggregate graduation rate will grow very slowly due to the increasing Hispanic share of the population.
That is from a new NBER working paper by
They are solving for the (crypto) equilibrium
Twenty-five people, including six minors, were charged in Paris over a spate of kidnappings and attempted abductions in France’s cryptocurrency world, said the city’s public prosecutor office on Saturday, May 31.
“Eighteen people have been placed in pre-trial detention, three have requested a deferred hearing and four have been placed under judicial supervision,” the public prosecutor said, with the suspects between 16 and 23 years old.
Here is the full story.
Italy facts of the day
About 156,000 Italians left the country last year for Germany, Spain, the UK and elsewhere, a 36.5 per cent increase over the number who emigrated in 2023.
At just under 191,000, the total number of people who left Italy in 2024 — including 35,000 long-term foreign residents, mainly Romanians returning home — was at the highest level in a quarter of a century, according to Italy’s official statistics agency, Istat.
Italy’s population decline is among the most acute in Europe, after decades of plummeting birth rates. At present, about a quarter of Italy’s 59mn people are over the age of 65, while just 12 per cent of the population are children aged 14 and under. The working age population is forecast to drop by another 5mn people by 2040.
Here is the full FT story.
Monday assorted links
Madrid’s Galería de las Colecciones Reales
Visitors don’t seem to know about this place yet, but it is one of the finest artistic venues in Spain. Taken from the royal collections and opened only a few years ago, it has one of the best displays of 16th century Brussels tapestries you will find, perhaps the best, beautifully hung with plenty of space. The paintings are from Goya, Velazquez, Melendez, Patinir, Mengs, Juan de Valdés, and others patronized by the Spanish rulers. Few bad pieces in the lot. There are also Goya tapestries, sometimes right next to associated Goya paintings. A splendid royal carriage.
This is perhaps my sixth (?) visit to Madrid, and the place never has felt better. Great for walking, and full of young people and small shops. It has absolutely displaced Barcelona as the leading city in Spain. A+ for both dining and art, and now it is the European capital of Latin America as well. It is no longer crazy to put it in the same league as Paris or Berlin, and these days feels more like a work in progress — in the good sense of that term — than either of those other places.