Germany fact of the day sentence to ponder
Champions of a European AI model should ask themselves if a European effort would be more effective than Meta, which this year will spend more on chips ($125 billion) than Germany spends on defense ($114 billion) and offer salaries of over $100 million to attract the best researchers, and is still failing to catch up.
Here is more from Pieter Garicano and Simon Grimm. Via Jesper.
Monday assorted links
1. A survey on slow Mexican economic growth.
2. Jason Furman on Social Security (NYT).
3. Markets in everything, customized water edition.
4. AI progress in Rio de Janeiro.
5. Satya Nadella does Oliver Williamson.
6. A shared feed of my guest appearances.
Republic of Ireland (China) fact of the day
Sam Enright emails me:
In the most recent census (2022), 1,017,437 people in Ireland were born abroad. Even if you classify people from Taiwan as “foreigners”, there are 845,697 + 157,886 = 1,003,583 immigrants to China. There are now more foreigners in Ireland than in China in absolute terms, despite having a population that is 260 times smaller.
Who Leads? Relative Age Effects on Social Capital
A fascinating paper and result:
This paper studies the causal effect of being the oldest within a school cohort on social capital. Using a fuzzy regression discontinuity design and data from Facebook, we find that boys who are older than their classmates make 11% more friends in high school. This social advantage is associated with leadership roles, with relatively older boys 42% more likely to become class president than their relatively younger peers. Men who were relatively older during childhood have larger social networks in adulthood, and disproportionately sort into management and entrepreneurship. Our findings suggest that small age differences in peer composition can have persistent effects on social and economic outcomes.
That is from Matthew Jacob of Harvard and Michael Bailey of Facebook. Via the excellent Kevin Lewis.
General-purpose large language models outperform specialized clinical AI tools on medical benchmarks
This result does not surprise me at all. Here is part of the abstract:
Frontier LLMs outperformed clinical AI tools in all three evaluations. Clinical AI tools performed comparably to auto-enabled Google Search AI Overview on the RCQ. These findings highlight the need for independent, real-world evaluation of AI tools before they enter clinical settings.
From Krithik Viswanath, et.al. As a side note, this (and the more general version of the point) is one big reason why some fairly large number of Emergent Ventures proposals are rejected rather quickly.
*The Pressure* (no spoilers)
A truly excellent movie, one of the best of the year. Specifically, it concerns the meteorological forecasts (!) leading up to the D-Day invasion. Thematically, it is about the differences between Americans and Brits, how bureaucracy operates, the nature of leadership, and the proper role of science in government. It is like an old-style Hollywood movie. Most of the action takes place in only a few rooms, and with superb dialogue and performances. Although you all know how D-Day turns out, the movie still generates suspense on some of the major plot points. Definitely recommended, here is the movie’s trailer.
Sunday assorted links
1. Chinese overtake Dominicans as NYC’s most numerous foreign-born group.
2. David Hockney embraced tech (NYT). And do not forget his writings, not to mention his persona and also his role in gay history and liberation. He was truly one of the great Englishmen, as he had been doing first-rate work since what, 1954?
3. Progress against lung cancer.
4. Average German date? And a six-minute video of a non-average German circus artist. And an eleven-year-old German on the handpan, without training.
5. Measuring how New Yorkers responded to their game 4 playoff victory. I have not seen data on game 5, though that was less of a surprise.
6. How The Bulwark is doing, and its economics (WSJ).
7. Become a telescope rancher those new service sector jobs (short video).
8. Bob Dylan (and others) on turning 80 (NYT). Dylan’s answer is clearly the best.
The Cultural War is a Civil War
Kevin Bryan riffs on on my post The Nationalization of American Science. He is rightfully incensed:
AT is right this is a red tape-filled science policy of “losers”. If you think “cut funds from DEI-driven professors in the small departments no one cares about” is more important than “make sure the world’s strongest fundamental science continues”, you’re an idiot.
And yes, this is also the policy of “right-wing JD-brain” folks. They haven’t worked in a lab. They don’t know how we got AI, and recent cancer breakthroughs, and on and on. It’s all culture war, all the time – just the right-wing equivalent of the worst left-wing habits.
One last thing: I *hate* the term “administration priorities” or “President’s priorities”. Totally Unamerican! The President *executes* the law created by Congress, who represent the people, and who see turnover every two years. Period. “Oh, but Democrats do this too!” Grow up!
Owning the libs may feel good today but please look just one move ahead in the game tree. When AOC controls the executive branch, she will inherit every tool Trump normalized. Look a few moves further and see the damage to American institutions.
The culture war is a civil war. If we don’t end it, American science will be collateral damage.
The bullish case for Brazil
Start with the most important number in economics, even though no one on Wall Street talks about it: calories per acre. Human civilization runs on food. Ten billion people will inhabit this planet by 2050. The amount of arable land is not growing. It is shrinking, every year, to urbanization, desertification, salinization, and topsoil erosion. The countries that can grow food at scale will be the most strategically valuable territories on earth. The countries with the best apps and the most PhDs will depend on the countries with the best dirt.
Brazil has more unused arable land than any country on earth. That sentence alone should stop every allocator in their tracks. It means that Brazil can approximately double its total cultivated area, without touching a single hectare of the Amazon, simply by converting degraded pasturelands in the Cerrado and other biomes into productive cropland using technology that already exists.
No other agricultural superpower has this headroom. The United States is fully utilized. China is losing farmland to urbanization at a rate that should terrify its central planners. India’s agricultural productivity gains are hitting diminishing returns against water stress and soil degradation. Europe is hemmed in by geography and regulation. Sub-Saharan Africa has theoretical potential, but lacks the roads, the ports, the legal frameworks, and the capital to exploit it within a generation.
Brazil is already the world’s largest net food exporter. It leads the world in soybeans, coffee, sugar, orange juice, beef, and poultry. It is the second-largest exporter of corn, pork, and ethanol, and recently surpassed the United States as the largest cotton exporter. Agribusiness generates approximately 25% of GDP and more than 40% of export revenue. And the agricultural sector has been growing productivity at 3-4% per year for two decades straight, driven by Embrapa’s tropical soil science, satellite-guided precision agriculture, and the industrialization of protein supply chains that stretch from feedlots in Mato Grosso to dinner tables in Shanghai.
A single farm in Mato Grosso can be more than twice the size of the state of Rhode Island. A literal fact. The Bom Futuro Group cultivates more than 700,000 hectares (roughly 2,700 square miles) of soybeans, corn, and cotton across 35 production units. This is farming at a scale that American and European investors cannot easily conceptualize, operating with GPS-guided machinery, drone monitoring, and soil analytics that rival anything in Iowa, but across an area that dwarfs it.
The post is interesting throughout and offers further points of interest.
Saturday assorted links
Sometimes it is hard to solve for the equilibrium
Probably you all know about this:
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
According to not yet confirmed but likely true reports, it was shown that model could be jailbroken. The released Mythos already restricted bio and “AI improvement” queries, rather strictly in fact, so now we are back to the model not being available.
Here are a few of the constraints on the U.S. government, not the only ones I might add:
1. It needs for the main companies to stay in business. On top of that, it wants their IPOs to go reasonably well. And it is now much harder for the top companies to recruit foreigners, which is a significant share of their highest quality workforce (Demis, Ilya, Andrej for a start). It is also much harder for the main companies to drum up foreign business in a credible and sustainble manner.
1b. How are American multinationals operating abroad supposed to use top systems, moving forward?
2. It wants to use model access as a tool of both hard and soft power, so model access has to be possible at some level. But it is very hard to control what foreign agents will do with their partial model access, when they get it in the ffuture.
3. The U.S. needs to stay ahead of China in the AI race.
4. The U.S. needs to issue restrictions that are actually enforceable, and “U.S. citizens only” does not fit that bill. Furthermore (markets in everything!) it is easy enough to hire a traitorous American to access tools of wrongdoing, or for matter it is not difficult to fake citizenship in various ways.
5. USG cannot nationalize these companies and then proceed to run them effectively.
6. Chinese and other open source models do in fact improve at some reasonable pace, even if they are right now considerably behind the best proprietary models.
Is the most likely scenario that the government hardens some of its own systems and takes some further precautions, and then allows Mythos to be rereleased? Perhaps with some additional safeguards?
Is there such a thing as a model that cannot be jailbroken at all? I doubt that.
So basically we will be replaying this scenario periodically over time, but with each time the companies and also the government in a weaker and more precarious position.
I am willing to reject the philosophy of “safetyism” and bite various associated bullets. As it stands, these actions will not succeed in making us safer, including for the reasons mentioned above. Our regulatory institutions, attitudes, and approaches simply are not well suited to an era of radical innovation.
In any case these events do not surprise me (they do surprise me in their immediate suddenness however), as this kind of approach is what governments have been about for a long time now, USG included or perhaps USG especially.
Rising in status: Leopold, Aesop, and also Mistral. AI nationalism. Proponents of slow take-off as the likely scenario. Reticent, quiet CEOs. As for China, will they rush into this opportunity, or are they at least as scared as we are?
How did Stanislaw Lem imagine advanced computer intelligence?
…GOLEM’s behavior is unpredictable. Sometimes it converses courteously with people, whereas on other occasions any attempt at contact misfires. GOLEM sometimes cracks jokes, too, though its sense of humor is fundamentally different from man’s. Much depends on its interlocutors. In exceptional casese GOLEM will show a certain interest in people who are talented in a particular way; it is intrigued, so to speak, not by mathematical aptitude — not even the greatest — but rather by interdisciplinary forms of talent; on several occasions it has predicted with uncanny accuracy achievements by young, as yet unknown, scientists in a field which it has it self indicated. (After a brief exchange it informed T. Vroedel, age twenty-two and then only a doctoral candidate, “You will become a computer,” which was supposed to mean, more o less, “You will become somebody.”)
That is from Lem’s Imaginary Magnitude, an extraordinary book in parts, most of all see his Golem IV section on how n AGI (our term, not his) is likely to behave.
Why is America less of a 24/7 society?
It’s deeply odd to me that America is a far less 24/7 hour society today than it was 10, 20, or even 30 years ago. I vividly remember friends from the UK back in 1996 marveling at the fact that in the mid-sized Indiana town where I went college it was possible to buy groceries, clothing, a lawn mower, a snow blower, Lego sets, and bow hunting gear at 3 AM on any given Tuesday of the year. That was peak American Empire, and it’s long gone.
That is from Christopher Kratovil. What are some hypotheses here? I see a few:
1. America is older. True, but this is hardly the main explanation for anything.
2. Due to increasing leisure time, fewer people want to work weird and long hours? Tighter labor markets and the Great Moderation contributed to this.
3. It is stores that are in decline. 24/7 activity has moved into the warehouse, the fulfillment center, the server farm, the delivery network, and the home.
3b. When you can do Doordash at 10:30 p.m., you do not need to go out for snacks at 3 a.m.
4. Shoplifting has become more common? If the drug stores have to lock up their wares in NYC, why should stores try to be open at 3 a.m., when presumably shoplifting risk is higher and the quantity of monitoring labor is lower?
5. Online entertainment is much better, so why go out late at night?
6. More work from home means people are not returning from their jobs at late hours and then wanting to buy things.
I would put most of my money on #3 and #5 — what do you think?
Safety and nation-building in Mexico
That is the topic of my latest Free Press column, here is one excerpt:
Consider the special nature of Mexican politics. First and foremost, Mexico is still not a mature nation-state. By one estimate, drug gangs may control as much as one-third of its territory. That might sound bizarre, but from the standpoint of Mexican history, it is not new or unusual.
Start with the 19th century. When Mexico won its independence from Spain in 1821, what we now call Central America joined the new country only briefly and then split off, even though that land was under the same Spanish jurisdiction. Those cultures and economies were not sufficiently unified to come along.
After independence, the state of Yucatán rebelled repeatedly, almost claiming its independence. In the 1840s, the U.S. declared war on Mexico and took away about half of its territory. Texas already had seceded to become an independent republic. In 1857, Mexico fought a civil war. The French invaded in 1861, and by 1864 they helped install a Habsburg, Maximilian, as emperor. Yet Maximilian never came close to controlling the entire country, and was quickly deposed and executed. The 1910 Mexican Revolution killed about 10 percent of the population by some estimates.
The rest of the 20th century was more peaceful, but much of Mexico never fell under unitary rule as did the U.S. and Western Europe. The more remote areas were mostly on their own, and they regarded the government as a potential oppressor rather than a savior. So when the drug trade heated up in Mexico in the 1990s as Colombian traffickers were partially thwarted, drug gangs were able to operate in many parts of Mexico with impunity. Eventually, they became the de facto rulers of those territories, supplying public goods such as general protection in addition to running their illegal businesses. All for a high price, of course, as extortion is still the ruling principle in those parts of the country. If you buy avocados from Mexico, for instance, there is a good chance that part of your money is going to pay tribute to drug gangs.
Another significant fact about Mexico is the size and power of its central government. It spends just short of 23 percent of its gross domestic product (GDP), relatively low for a country of its level of development. By contrast, Brazil, which has roughly comparable living standards, has a central government that spends over 32 percent of that country’s GDP. If the Brazilian government is too large, Mexico’s is too weak and too small, most of all because Mexico cannot beat back its drug gangs by brute force or preempt them in the first place.
Mexico as a topic will never become obsolete, not for the United States at least.
Friday assorted links
1. Is there too much free parking in NYC? (NYT)
2. New Malcolm Gladwell book forthcoming on violence in America. Ready for pre-order.
3. Manufacturing requirements are killing gene and cell therapy.
5. What went wrong with German trains? (FT)
6. Why do people wander in a counterclockwise direction? (NYT)
7. Seb Krier: “Over the past few months I’ve been working on a very exciting project: a new $10m fund for research on multi-agent multi-principal AGI safety! Instead of focusing on single agent alignment and centralized control, we’re looking to support research focusing on multi-agent settings, mechanism design, cooperative AI, and coordination problems.”