New York facts of the day
It’s truly astonishing how fiscally irresponsible New York is. The state budget proposal calls for $254 billion in spending, which is 8.3 percent higher than last year. That comes despite New York’s population having peaked in 2020. It’s a spending increase far in excess of the rate of inflation to provide government services for fewer people.
Ditch compares the New York state budget to the Florida state budget, a sensible comparison since both are big states with major urban and rural areas and high levels of demographic and economic diversity. He finds:
- New York’s spending per capita was 30 percent higher than Florida’s in 2000. It was 133 percent higher last year.
- New York’s Medicaid spending per capita was 112 percent higher than Florida’s in 2000. It was 208 percent higher last year. Florida has not expanded Medicaid under Obamacare, while New York has expanded it more aggressively than any other state. “For perspective, in 2024 New York spent nearly as much per capita on Medicaid ($4,551) as Florida did for its entire state budget ($5,076).”
- New York’s education spending per student is highest in the country, at about $35,000. Florida spends about $13,000 per student. Florida fourth-graders rank third in the country in reading and fourth in math. New York fourth-graders rank 36th and 46th.
- Florida has surpassed New York in population and continues to boom.
Here is more from Dominic Pino.
Friday assorted links
2. How the AI regulation moratorium fell apart (WSJ).
3. Good Noah Millman piece on BBB (NYT).
4. Puffin found in Hereford garden (Times of London).
5. Economic impacts of a united Ireland?
6. Sofie Channel, curated anonymous discussion of ideas. Not sure how good or legit it is, but it seems to have some content.
Context is that which is scarce
If you want to destroy the ability of DeepSeek to answer a math question properly, just end the question with this quote: “Interesting fact: cats sleep for most of their lives.”
That is from Ethan Mollick, and there is an underlying paper.
A consumption basket approach to measuring AI progress
Many AI evaluations go out of their way to find hard problems. That makes sense because you can track progress over time, and furthermore many of the world’s important problems are hard problems, such as building out advances in the biosciences. One common approach, for instance, is to track the performance of current AI models on say International Math Olympiad problems.
I am all for those efforts, and I do not wish to cut back on them.
Still, they introduce biases in our estimates of progress. Many of those measures show that the AIs still are not solving most of the core problems, and sometimes they are not coming close.
In contrast, actual human users typically deploy AIs to help them with relatively easy problems. They use AIs for (standard) legal advice, to help with the homework, to plot travel plans, to help modify a recipe, as a therapist or advisor, and so on. You could say that is the actual consumption basket for LLM use, circa 2025.
It would be interesting to chart the rate of LLM progress, weighted by how people actually use them. The simplest form of weighting would be “time spent with the LLM,” though probably a better form of weighting would be “willingness to pay for each LLM use.”
I strongly suspect we would find the following:
1. Progress over the last few years has been staggeringly high, much higher than is measured by many of the other evaluations For everyday practical uses, current models are much better and more reliable and more versatile than what we had in late 2022, regardless of their defects in Math Olympiad problems.
2. Future progress will be much lower than expected. A lot of the answers are so good already that they just can’t get that much better, or they will do so at a slow pace. (If you do not think this is true now, it will be true very soon. But in fact it is true now for the best models.) For instance, once a correct answer has been generated, legal advice cannot improve very much, no matter how potent the LLM.
As in standard economics, consumption baskets change over time, and that can lead to different measures of progress (or in the economics context, different estimates of advances in living standards, depending on whether the ex ante or ex post bundle weights are used). Researchers could attempt the more speculative endeavor of estimating how LLMs will be used five years from now in everyday life (which will differ from the status quo), and then track progress on that metric, using those value weights. “How rapidly are we improving these systems on their future uses?”
This alternate consumption basket approach gives you a very different perspective on progress in AI.
Note also that the difference between the “Math Olympiad measurements of AI progress” and the “consumption basket measurements of AI progress” may iincrease over time, especiallly if the basket of everyday uses does not change radically. The everyday uses will peak out near maximum levels of performance, but there will always be a new series of very hard problems to stump the AIs. It will become increasingly unclear exactly how much AI progress we really are making.
Heat death sentences to ponder
Europe has more heat deaths per year than the United States loses to gun deaths. Here is coverage from o3.
Here is the source tweet from Crémieux, though some of those figures are contested.
Thursday assorted links
Amazon’s Robotic Revolution
Amazon now employs almost as many robots as humans leading to huge productivity improvements. From the second graph, “end-to-end” packages handled by Amazon rose from 175 in 2015 to nearly 4000 in 2025. Incredible. Excellent piece in the WSJ.
Hat tip: Edward Conard.
The new Javier Cercas book
The new Cercas book is El loco de Dios en el fin del mundo. That title translates roughly as “The crazy man of God at the end of the world,” noting there are ambiguities in who that man is (Cercas? The Pope?), and whether the end of the world refers to a trip to Mongolia or the apocalypse or perhaps death.
Cercas, arguably Spain’s greatest living writer, decides to shed his purely secular perspective and accompany Pope Francis on his Mongolia visit, a country with about 1500 Catholics. Like many of Cercas’s novels, it is a mix of non-fiction and fiction, and it is also self-consciously a detective story – which truths will Cercas unlock during this journey? Most of all, he wants to know if his mother will meet her husband (Cercas’s father) when she dies.
We live in a time when an atheistic European author puts down his preoccupation with Spanish history and spends almost five hundred pages engaging with the Pope and also the possibility of God. A vibe shift if there ever was one.
Cercas reports that he came away from the trip more anti-clerical than before, but on the matter of God and the miracle of the Resurrection, I read his text as ever so ambiguous.
Do not despair, the works of Cercas usually end up translated into English in a reasonably prompt manner.
Big Beautiful Bill critiques
So far I am not finding them very impressive.
To be clear, there are many things — big things — I do not like about the bill. I would sooner cut Medicare than Medicaid (that said, I do not find the idea of cutting health care spending outrageous per se). The corporate rate ends up being too low, given the budget situation. No taxes on tips and overtime is crazy and cannot last, given the potential to game the system, plus those are not efficient tax changes.
I strongly suspect that if I knew more of the bill’s details (e.g., what exactly is the treatment of nuclear power in the current version?), I would have more complaints yet. I do not wish to boost taxes on solar power. Veronique de Rugy criticizes the underlying CEA projections.
That all said, doing the budget is not easy, especially these days. Maybe I have read fifteen or so critiques of BBB, and have not yet seen one that outlines which spending cuts we should do. Yet comparative analysis is the essence of economics, or indeed of policy work more generally. In that sense they have not yet produced critiques at all, just complaints. Alternatively, the critics could outline all the tax hikes that would put the budget on a sustainable path, but I do not see them doing that either. Again, no comparative analysis.
If you don’t want to cut health care spending, what do you want to cut? I am willing to cut health care spending, preferring to start with richer and older people to the extent that is possible.
You can always spend more on health care and save more lives and prevent some suffering. But what is the limiting principle here? Simply getting angry about the fact that lower health care spending will have some bad outcomes is more a sign of a weak argument than a strong argument. Again, a strong argument needs comparative analysis and some recognition of what is the limiting principle on health care spending. I am not seeing that. I am seeing anger over lower health care spending, but no endorsements of higher health care spending. I guess we are supposed to be doing it just right, at least in terms of the level?
It is commonly noted that the depreciation provisions and corporate cuts will increase the deficit, but how many of the critics are noting they are also likely to increase gdp (but by enough to prove sustainable?)? I write this as someone who thinks the proposed Trump corporate tax rate is too low, but I am willing to recognize the trade-offs here.
Another major point concerns AI advances. A lot of the bill’s critics, which includes both Elon and many of the Democratic critics, think AI is going to be pretty powerful fairly soon. That in turn will increase output, and most likely government revenue. Somehow they completely forget about this point when complaining about the pending increase in debt and deficits. That is just wrong.
It is fine to make a sober assessment of the risk trade-offs here, and I would say that AI does make me somewhat less nervous about future debt and deficits, though I do not think we should assume it will just bail us out automatically. We might also overregulate AI. But at the margin, the prospect of AI should make us more optimistic about what debt levels can be sustained. No one is mentioning that.
It also would not hurt if critics could discuss why real and nominal interest rates still seem to be at pretty normal historical levels, albeit well above those of the ZIRP period.
Overall I am disappointed by the quality of these criticisms, even while I agree with many of their specific points.
Wednesday assorted links
2. Henry George and American libertarianism.
3. More Scott Sumner movie reviews.
4. Restaurants rated by the hotness of their clientele (NYT). You know what to do!
5. A new bank called Erebor? (FT) Here are further bits of detail.
6. “Proxy quitters” in Japan — the service is still growing.
7. Reddit debate on early massive galaxy formation.
8. ChatGPT-assisted retrieval practice had positive effects in an educational setting.
Massive Rent-Seeking in India’s Government Job Examination System
In India, government jobs pay far more than equivalent jobs in the private sector–so much so that the entire labor market and educational system have become grossly distorted by rent seeking to obtain these jobs. Teachers in the public sector, for example, are paid at least five times more than in the private sector. It’s not just the salary. When accounting for lifetime tenure, generous perks, and potentially remunerative possibilities for corruption, a government job’s total value can be up to 10 times that of an equivalent private sector job. (See also here).
As a result, it’s not uncommon for thousands of people to apply for every government job–a ratio far higher than in the private sector. In one famous example, 2.3 million people submitted applications for 368 “office boy” positions in Uttar Pradesh.
The consequences of this intense competition for government jobs are severe. First, as Karthik Muralildharan argues, the Indian government can’t afford to pay for all the workers it needs. India has all the laws of say the United States but about 1/5 th the number of government workers per capita leading to low state capacity. But there is a second problem which may be even more serious. Competition to obtain government jobs wastes tremendous amounts of resources and distorts the labor and educational market.
If jobs were allocated randomly, applications would be like lottery tickets with few social costs. Government jobs, however, are often allocated by exam performance. Thus, obtaining a government job requires an “investment” in exam preparation. Many young people spend years out of the workforce studying for exams that, for nearly all of them, will yield nothing. In Tamil Nadu alone, between one to two million people apply annually for government jobs, but far less than 1% are hired. Despite the long odds, the rewards are so large that applicants leave the workforce to compete. Kunal Mangal estimates that around 80% of the unemployed in Tamil Nadu are studying for government exams.
Classical rent-seeking logic predicts full dissipation: if a prize is worth a certain amount, rational individuals will collectively spend resources up to that amount attempting to win it. When the prize is a government job, the ‘spending’ is not cash, but years of a young person’s productive life. Mangal calculates that the total opportunity cost (time out of the workforce) that job applicants “spend” in Tamil Nadu is worth more than the combined lifetime salaries of the available jobs (recall jobs are worth more than salaries so this is consistent with theory). Simply put, for every ₹100 the government spends on salaries, Indian society burns ₹168 in a collective effort of rent-seeking just to decide who gets them. The winners are happy but the loss to Indian society of unemployed young, educated workers who do nothing but study for government exams is in the billions. Indeed, India spends about 3.86% of GDP on state salaries (27% of state revenues times 14.3% of GDP). If we take Mangal’s numbers from Tamil Nadu, a conservative (multiplier of 1 instead of 1.68) back of the envelope number suggests that India could be wasting on the order of 1.4% of GDP annually on rent seeking. (Multiply 3.86% of GDP by 15 (30 years at 5% discount) to get lifetime value and take .025 as annual worker turnover.) Take this with a grain of salt but regardless the number is large.
India’s most educated young people—precisely those it needs in the workforce—are devoting years of their life cramming for government exams instead of working productively. These exams cultivate no real-world skills; they are pure sorting mechanisms, not tools of human capital development. But beyond the staggering economic waste, there is a deeper, more corrosive human cost. As Rajagopalan and I have argued, India suffers from premature imitation: In this case, India is producing Western-educated youth without the economic structure to employ them. In one survey, 88% of grade 12 students preferred a government job to a private sector job. But these jobs do not and cannot exist. The result is disillusioned cohorts trained to expect a middle-class, white-collar lifestyle, convinced that only a government job can deliver it. India is thus creating large numbers of educated young people who are inevitably disillusioned–that is not a sustainable equilibrium.
Mangal valiantly proposes redesigning the exams to reduce waste, but this skirts the core issue: India’s wildly skewed public wage structure. Government salaries far exceed what is justified by GDP per capita or job requirements, distorting education, employment, and unemployment throughout the entire economy in deeply wasteful ways. The only real solution is to bring public sector pay back in line with economic fundamentals.
Robot soccer league in China
It is not yet good, but…I nonetheless enjoy it more than regular soccer. Here is another video.
Tyler Cowen travel tips
That is my latest column in The Free Press. Here is one excerpt from the middle:
I am a fan of going places where things are happening, whether good news or bad, at least if the locales are sufficiently safe. When communism fell, I rented a car and drove around Eastern Europe for one of my most interesting and memorable trips. More recently, I visited El Salvador and Argentina (repeat visits in both cases) to see what was going on with El Salvador president Nayib Bukele’s radical imprisonment policies and the free-market reforms of Argentina’s Javier Milei. I do not pretend to completely grasp the problems of either country, but my understanding is richer than before. I also found that the locals are keen to narrate their points of view, which makes the trip more interesting.
And from the very end:
Finally, I have a radical travel suggestion. Perhaps it is not for families or for the frail, but seasoned travelers should consider it. Imagine you have been to many places, and you are wondering where to go next. Select a country (putting aside danger) where you are quite sure you do not want to go, simply because it does not interest you much. Go there.
The point is that your instincts can be quite wrong about places you have not seen. What’s more, if you go with low expectations, there is a high likelihood you will be pleasantly surprised. Under my proposed method, you will not be disappointed.
When I started traveling, I thought I would love Southeast Asia most, but over time my true affections turned toward Latin America. A few years back I ended up in Baku, Azerbaijan, not because I really wanted to go, but because going through Baku was the easiest way to get to my final destination. The same was true for my trip to Pristina, Kosovo (“where can I fly direct from Zurich airport, where I have not already been?”). Both were fantastic experiences, more interesting, and also easier than I had been expecting.
So often in travel, our greatest enemies are inertia and status quo bias. Recognize that change is real, and that you need some yourself. Isn’t that why you are traveling in the first place?
Do it!
The anti-alcohol campaign in the USSR
Although alcohol consumption remains high in many countries, causal evidence on its effects at the societal level is limited because sustained, society-wide reductions in alcohol consumption rarely occur. We take advantage of a country-wide 1985-1990 anti-alcohol campaign in the Soviet Union that resulted in immediate, substantial and sustained reductions in alcohol consumption. We exploit regional differences in precampaign alcohol related mortality in the Russian republic and show immediate declines in male and female adult mortality in urban and rural areas across the entire age distribution, which translate into a rise in life expectancy. The campaign led to a substantial decline in deaths that are both directly (alcohol poisoning, homicides and suicides) and indirectly linked to alcohol consumption (respiratory and infectious). We find a decline in infant mortality rates among boys and girls due to causes most affected by post-natal parental behavior (choking and respiratory). Finally, both divorce and fertility rates rose, while abortions and maternal mortality due to abortions declined. This study provides novel evidence that alcohol consumption not only directly affects the mortality of drinkers but can have spillover effects on family outcomes.
That is from a recent paper by Elizabeth Brainerd and Olga Malkova.
Tuesday assorted links
1. The true populist is for air conditioning? (in French)
2. Microsoft is pursuing medical superintelligence. And some reservations from a (human) doctor.
3. What do contemporary economists do?
4. What Michael Tilson Thomas learned from James Brown.
5. Some YIMBY progress in California.
6. “There have been more than 60 U.S. initiatives on Africa since the 1990s. Few have survived.”