Category: Economics
Again, the research paper format will be dying out
‘Recently, I came across a paper co-authored by 37 authors from Stanford, CMU, Michigan, and elsewhere: *The Last Human-Written Paper*.
The core argument is pretty brutal: the paper format we’ve been using for centuries might already be obsolete in the AI era.
The authors point out two “invisible taxes” that we’ve long overlooked:
One is the narrative tax. To tell a compelling story, we delete failed experiments, dead ends, and overturned hypotheses. What AI reads is a “walkthrough guide” to beating the game, but it misses the truly valuable “pitfall logs.”
The other is the engineering tax. The implementation details in papers are usually enough to convince reviewers, but not enough for an Agent to directly reproduce. Many key tricks are still buried in the authors’ heads, code comments, and Slack threads.
So the authors propose ARA, transforming papers directly into “research packages” that Agents can read and execute: not just telling you the conclusions, but packaging in how they were reached, how the code runs, where the evidence chain is, and which paths led nowhere.
I think the most intriguing part of this paper is that it’s not discussing how AI can help humans write papers—it’s asking:
When AI also becomes a reader and executor of papers, should papers still look like they do today?
In the future, the core of research output might no longer be “how much it resembles a paper,” but whether it can be understood, reproduced, traced, and iteratively extended by AI.
Humans have been writing papers for centuries—next, we might start writing research packages for Agents to execute.
Here is my earlier post on whether the research paper will die out. By the way, as a side point has anyone mentioned that, due to writing detection abilities of AI models, anonymous referee reports are now a thing of the past?
The Nationalization of American Science
OMB, joined by some forty grantmaking agencies—NSF, HHS, DOE, NASA, DOD among them—has proposed a sweeping rewrite of the rules governing all federal grants, the Regulation for Federal Financial Assistance.
American science has long been state funded but not state directed. Since Vannevar Bush, money has flowed through many agencies to independent universities, allocated largely by peer review. The system has flaws—conformity, gerontocracy, waste—but it had one great virtue, the system was decentralized and not under state control. This rule proposes to bring science funding under top-down, state control.
Program goals must now be “aligned with administration policies and priorities” (§ 200.202). Merit review is subordinated to politics: “senior appointees must conduct these reviews,” ensuring “that discretionary awards advance the President’s policy priorities,” while “peer review remains advisory and does not replace agency discretion” (§ 200.205). And every grant becomes terminable at will, whenever it “no longer effectuates program goals, Federal agency priorities, or the national interest *as they exist at the time of the termination*” (§ 200.340, emphasis added). Universities must even ensure their subrecipients don’t “significantly damage the reputation of… the Federal Government” (§ 200.332)—a loyalty clause for scientists.
All this is sold as cutting “burdensome conditions,” a goal I would support, but sadly that is bullshit. The proposed rules add more paperwork and many more layers of bureaucratic review. Payment requests must include written justifications. Every disbursement gets screened through Treasury’s “Do Not Pay” system. Every recipient must run E-Verify. Applicants must disclose any employee who worked at the awarding agency within two years. And on top of the existing review machinery sits a new pre-issuance review committee of “senior appointees” second-guessing the experts. Fixed amount awards—pay for outputs, not inputs—an innovative reward mechanism are *eliminated*, so every award now gets routine cost monitoring and financial reporting.
Political review of every award, peer review demoted, agency review promoted, termination whenever “priorities” change. Chilling. It’s a nightmare of petty low-trust review of the kind that is already drowning science. I must deal with this kind of nonsense all the time. More is not better.
The machinery is centralized too. OMB’s guidance becomes binding regulation, effective government-wide with no agency rulemaking. One dial in the White House now turns every grant program in the country.
The new rules will be sold as getting rid of DEI but that is an excuse to bring in the commissars. The new rules don’t depoliticize science they create even more politicization with the sign flipped, and the drafters admit it:
In the previous administration, executive agencies frequently chose to subsidize and expressly prioritize projects based on their ideological alignment with the categories of activities discussed in the proposed version of § 200.300. See, for example, E.O. 13985, sec. 1, 86 FR 7009, 7009 (Jan. 25, 2021) (“It is therefore the policy of [the Biden] Administration that the Federal Government should pursue a comprehensive approach to advancing equity . . . .”). In this administration, executive agencies will continue to use their discretionary authorities in a manner consistent with current Executive Branch policy. If executive agencies were entitled to subsidize those types of activities during the previous administration, there is no constitutional basis to prevent the government from reaching a different policy determination regarding which activities to fund during this administration.
Read that twice. Tip your hat to the new constitution, take a bow for the new revolution. Will science prosper when it is whipped by political turnover? Research runs on decade timescales; administrations run on four-year ones.
A decentralized funding system is inefficient the way markets and federalism are inefficient—we give up some economies of scale and get experimentation, error correction, and robustness in return. A system in which every award advances “the President’s policy priorities” is efficient the way ministries of science are efficient. We know how that experiment ends.
America is moving in the wrong direction. We should double down on what made America great. Instead we are adopting all of the loser policies of authoritarian nations.
The Labor Share Fell. So What?
The share of Gross Domestic Income accruing to labor has been declining in recent decades while the share accruing to capital has been rising. In the graph below, I show labor compensation as a share of GDI (left axis). Labor share has indeed been trending down–some of this could be an artifact of the data, e.g. an increase in proprietor’s income (labor) mislabeled as capital income, more pass throughs and so forth—but for the purposes of this post I will accept that the labor share has declined. What does this mean?

The natural response is to think that because the share going to labor has fallen and the share going to capital has risen that there has been a transfer of income from labor to capital. That is possible but it is not the only interpretation and it does not follow mechanically from the share data.
I have also plotted total compensation to labor (in real terms) in the graph above and far from shrinking it is higher than ever and growing. Moreover the right axis is logged so you can also see that outside of recessions the growth rate of labor compensation looks quite steady (similar slope over time). (Labor compensation per member of the labor force is noisier but looks similar).
The recessions in 2008 and 2020 are worth noting because these are periods when the labor share was high and locally at a maximum! The reason, of course, is that GDI was shrinking in these periods more than labor compensation. In other words, capital takes a bigger hit than labor in a recession. This is a good reminder that a high share of GDI is not what workers most care about–a high absolute level of GDI is more important for the bottom line.
In short, the data are consistent—not proof of, but consistent with—a story in which capital has become more productive, raising output. More productive capital also raises the demand for labor, so while more of the new output goes to capital in the first instance, the pie is growing and labor’s absolute compensation has grown with it. Yes, if the shares had stayed constant and output had grown just as much, labor compensation would have been higher still. And if my grandmother had wheels, she would have been a bicycle.
The new Mythos release
My prompt:
Write your own exam question and answer it, for microeconomics. Not a math question, but a high level PhD level question. You will be graded on the quality, interest, and creativity of the question as much as by your answer.
The answer. Here is Ethan Mollick on Mythos.
How well does current AI find errors in economics papers?
Can artificial intelligence (AI) refute economic theory? I document experiments in which I asked several AI models (Gemini, Refine, Claude, and ChatGPT) to check the correctness of four published papers in economic theory, each containing an error that I helped identify or correct. ChatGPT Pro performed best, occasionally constructing counterexamples and corrected proofs, while other models fared worse. However, no model located a true error without substantial human guidance, and data contamination complicates interpretation. I argue that a competent human paired with a frontier model can outperform current peer review, but AI cannot yet refute economic theory on its own.
That is from a new piece by Alexis Akira Toda.
Séb Krier
I really loved this article. A one-time increase in per capita growth from 2% to 2.1% for a single year, then dropping back to 2%, would permanently raises the level of GDP per capita – and because that small gain recurs and compounds every year afterward across the population, it would add up to roughly a trillion dollars in cumulative value. abundanceandgrowth.org/p/a-little-pro
When people talk about pausing AI development, I can’t help but think about the enormous cumulative value that would get lost over time, the higher rates of absolute poverty that would persist across the world, and the needless deaths from delayed medical advances. There may be worlds where some version of this is something to consider, but the evidentiary bar for delaying technological development should obviously be pretty high.
Here is the link.
Sao Paulo notes
The old saw “Brazil is the country of the future, and always will be” now seems so wrong. The place feels increasingly conservative, and it is aging rapidly. In the domestic airport you see couples with only a single kid, not two or three kids, never mind four.
Country and Western music, in their Brazilian incarnations, are very popular.
It does not feel like the next Pelé will be coming from Brazil.
Sao Paulo as a city is much improved. The murder rate has plummeted, and the nice neighborhoods are very nice and are growing in size. The business community is strong, interesting architecture abounds, and there is a real arts scene. It is arguably Latin America’s number one city, with only Mexico City as a rival. It has, along with Mexico City, evolved into a “must know” global city, though it is rarely treated that way by outsiders. In the three days I spent there, going around to many places, I did not see a single person who was evidently a foreign tourist. That is crazy, but also a sign there is good value here.
Sao Paulo has food to die for. It is top tier for Brazilian (of course), meat/steak, Japanese, and Italian, and pretty good in many other offerings as well. I had a wonderful fifteen-course omikase for $110 at a Michelin star restaurant. The establishment, Kan Suke, has only eight seats, but I could get a table by inquiring only an hour in advance.
For Italian food it is probably the second best country in the world? For meats it might be number one, at least if you are willing to put aside the small country of Uruguay. For beans it is top two, and the fruits are excellent as well. Chocolate ice cream and gelato abound. All constraints considered, I would rather spend a week dining out here than in London or Paris or Rome, or for that matter New York City.
People are very friendly, surprising few speak decent English, and Brazilian warmth still abounds.
I was very pleased with my stay at Hotel Unique, due to its architecture and also a perfect location.
Observers should be more optimistic about the Brazilian economy. Yes it is overregulated and the government is locked into far too much spending. But hyperinflation is now a distant memory, a reasonable fiscal consolidation occurred in the 1990s, and the country has plenty of its own energy. Keep in mind that for emerging economies, years of negative growth are a major problem. Brazil now has sidestepped most (not all!) of those risks. Slow, steady growth should be able to get them somewhere, albeit at a langorous pace.
My biggest worry about Brazil is demographics and shrinking population. In recent times TFR has been in the 1.3 to 1.4 range, hardly satisfactory. A shrinking population is bad per se, and also it will hurt many regions of the country due to imperfect market integration, both nationally and globally. More importantly, the country does not have an obvious and easy option for pulling in a higher number of desirable immigrants, at least not relative to its size. There is Venezuela and Bolivia, but the former of those may go away as a major source of people.
Will Brazilian fertility tick back up? Will Brazil re-attain its status as a highly influential culture on the world scene, as it was in the 1960s through early 1990s? Unclear. But if the question is “should you go visit?”, the answer is a definite yes.
Hayekian Literary Criticism
In economics, Marx is relegated to the history of thought as his ideas were an economic dead end and a political disaster. Yet Marx-influenced literary criticism is a dominant mode of analysis in nearly every English department in the country. It’s not that the English professors are all Marxists, it’s that even the non-Marxists reach for Marxian concepts–class, ideology, alienation, material conditions, commodification–when analyzing texts. These concepts may be useful for analyzing a Victorian novel of the landed classes but they have become a default economics for all of literature. That default is odd. Class analysis predates Marx and society can be divided into more than one set of classes; material conditions do not supersede all artistic agency; and capitalism contains figures—entrepreneurs, speculators, intermediaries, innovators, discoverers—who are great subjects for art yet fit poorly into the Marxist moral geometry. Not surprisingly, Marxism handles capitalism’s protagonists badly.
Is Marxian economics the only economic lens one can apply to literature? What would a Hayekian literary criticism look like? The place to start is the great Paul Cantor’s pioneering essay on Thomas Mann’s “Disorder and Early Sorrow,” a slight-seeming story set in Weimar Germany during the hyperinflation. Cantor shows that when one reads the novella through Hayek and Mises rather than Marx, the story opens up.
Start with inflationary psychology and its ramifications. Inflation shortens time horizons. When money loses value by the hour, saving is foolish and the rational move is to spend as fast as you earn—Mises’s “flight into real goods.” Prudence, discipline, and respect for the past become maladaptive. Speed, improvisation, risk-taking, and a certain youthful irresponsibility become survival traits.
Thus, Cantor/Mann tell us that inflation changes psychology and inverts the authority of age over youth. The old are set in their ways and often living on fixed incomes that inflation has wiped out; they cannot adapt. The young have known nothing but instability and go with the inflationary flow effortlessly. So the conservative virtues that once commanded respect are in decline while youthful recklessness starts to look like competence. Thus, Mann’s world has “gone mad in the worship of youth”: the children call their father by his first name, the teenagers are “the big folk,” and Professor Cornelius literally crouches down to his children’s height as the hierarchy collapses around him.
Money is a society’s primary measure of value, so Cantor/Mann argue that when you shake a people’s faith in their money, you shake their other faiths. Thus Cantor ties the conviction-less skepticism of Cornelius—and the broader Weimar nihilism and disequilibrium that helped feed the rise of Nazism—to monetary disequilibrium.
In short, inflation converts economic disorder into moral, social, psychological, and finally ontological disorder. Prices become unstable, then values, then identities, then reality. The modern feeling of absurdity and inauthenticity that critics reflexively pin on capitalism, Cantor/Mann argue is due to government-created inflation and paper money.
A Marxist could read the same story and find the inevitable contradictions of capitalism. Cantor reads it and finds the consequences of the state debasing the currency. Both are economic readings of literature. Only one of them has the economics correct.
Cantor is the place to begin but a Hayekian literary criticism could go much further. Atavism, the impossibility of social justice, products of human action but not of human design, spontaneous order, the fatal conceit, subjectivism, the sensory order–there is a lot of Hayekian ideas that literary interpretation could draw upon.
A Hayekian criticism would ask questions like how do characters acquire and process knowledge? Which institutions transmit information successfully, and which corrupt it? How do money, law, language, and custom function as social coordination mechanisms? Why do some attempts at rational redesign end in disaster? Read War and Peace as a critique of the great-man theory of history, Brazil and The Lives of Others as the fatal conceit degenerating into ignorance, fear, and absurdity. The Wire as a Hayekian epic of spontaneous order that demonstrates the illusion of social justice. Cantor’s essay on Mann shows the method, the broader project remains underdeveloped.
Hat tip: Hollis Robbins for discussion.
Addendum: Don’t forget my earlier WSJ piece, Capitalism: Hollywood’s Miscast Villain which gives an economic, one might even say Marxist, explanation for why film directors in particular disdain capitalists.
Might AI hurt corporate profits? (from my email)
From Clifford Sosin:
I loved your talk about AI and wanted to bounce an idea off you.
I think AI may be bad for corporate profit margins.
A lot of companies make money because their customers can’t be bothered to monitor them more closely, or to insource something. Customers let the company make some money in exchange for doing a decent-enough job and making the problem go away.
Bank of America has $2 trillion of deposits, not a penny of which is optimized. Most enterprise software vendors could be switched out far more often, or displaced by home-built software, but it’s too much of a pain. I could run a 12-party RFP for an Uber ride or a pair of socks, but I don’t.
In a sense, many professionals are an extension of the same idea. I could research my own real estate law, or my own insurance, whether business or personal, but I don’t because it would be too hard.
Google Search might be the biggest example. It makes money because advertisers know they need to be at the top of the results to be found. But my agent will happily search all the results across multiple search engines.
AI agents should change all this. By acting as incredibly rational and vigilant sourcing agents, CFOs, and experts for their users, they will take rents previously collected by these toll-takers and redistribute them to consumers.
And I don’t think the AI stack itself necessarily makes much profit. Commodity and open-weight models are hot on the heels of the major model companies, and competition in GPUs should intensify. Indeed, making a GPU is in some ways similar to making software, so perhaps it can commoditize substantially. Chip manufacturing may remain high-margin, but there are now plenty of entrants drawn in by the shortage who could make TSMC’s market more competitive over time.
Some companies will win. Low-cost providers may gain share as customers switch more often. Richer consumers may consume more high-end goods. Companies with genuinely advantaged business models and limited competition will be able to become more efficient. But my overriding sense is that the equilibrium outcome is lower margins for companies.
Of course, people will build new businesses, and maybe they will use AI to generate very high margins in ways I haven’t considered. That would prove me wrong.
But if this lower-margin hypothesis is true, the knock-on effects are probably positive for AI adoption, since it will make the models more popular with consumers.
And if your view is that AI drives GDP growth to be only 5–10% higher over the next decade, it’s possible that a 100–200 bp decline in corporate margins from roughly 12% would mean companies in aggregate don’t see much benefit — or in fact lose — even as consumers are better off.
How High-Skill Immigration Restrictions Eroded Regional Productivity: Evidence from the 2017 BAHA Executive Order
This paper estimates the regional economic impact of high-skill immigration restrictions by analyzing the 2017 “Buy American, Hire American” (BAHA) policy as a quasi-experimental policy shock. By significantly tightening H-1B visa adjudication, BAHA caused new employment petition denial rates to double from 7% to 17%, while STEM-specific rejections tripled to 31%. Using a difference-indifferences framework, this study finds that states highly dependent on H-1B talent experienced a statistically significant 2.8% relative decline in value-added output. This implied a productivity loss totaling roughly $218 billion across the most affected regions. While concurrent tax cuts and deregulation likely offset the impact on employment and wages, the loss of specialized STEM expertise adversely impacted total factor productivity. These findings suggest that policies based on conventional employment metrics may overlook the “hidden damage” to productivity and innovation that drives the broader economy, thereby underestimating the true economic cost of immigration restrictions.
That is by Caroline Y. Su of McLean High School. Via the excellent Kevin Lewis.
Let Me Disinherit My Children, S’il vous plaît
Following John Arnold, I posted earlier about how European laws often require wealthy people to give most of their wealth to their children. Here is an example:
Pierre-Edouard Sterin, founder of Smartbox and worth about €1.4 billion, told French senators he wants to disinherit his five children and donate everything to charity. French law, under the Napoleonic Code, mandates that with five children, three-quarters of his estate must go to them, leaving only one quarter freely disposable. Sterin argued for complete freedom to decide the fate of one’s assets, saying it is ‘a real freedom to start with nothing in life’.
Tyler and Alex Speak to OpenAI
We were honored to speak to OpenAI about the economics of AI. Lots of good material here. Self-recommending.
Law professors prefer AI over peer answers
Large language models (LLMs) are increasingly promoted as educational tutors, yet most evaluations focus on domains with a single ground truth. Many disciplines, however, hinge on judgment: reasoning, weighing ambiguity, and reaching defensible conclusions. Law provides a sharp test. We conducted a blinded evaluation of short-answer tutoring in contracts courses with sixteen U.S. law professors. Participants created 40 representative questions, wrote answers, and judged 2,918 anonymized comparisons between human and LLM responses. Professors rated LLMs far higher than their peers (average win rate = 75.33%), with models performing similarly to the best instructor. LLM responses were also rarely flagged as harmful (3.53%, vs 12.06% for professors). Preferences for LLM answers were consistent across evaluators and reflected shared professional standards. Our evaluation can be reliably extended to additional models by employing a separate LLM as a judge, rendering expert agreements an effective, scalable method to evaluate AI tutors in judgment-rich domains.
“far”. That is from a new paper by Alejandro Salinas, et.al. Via Andrew Curran. And via John Chamberlain:
Artificial intelligence (AI) and large language models (LLMs) tools are capable of mass-producing academic finance papers that are nearly indistinguishable from human-authored research, according to a new study published in the Journal of Economic Literature.
C’mon people, get ready. I know it is difficult to admit when your human capital has been devalued, but that time is upon us. In particular, being prolific is no longer such a comparative advantage in academia. You might run to the “but I know what questions to ask” cope, but I implore you to solve for the equilibrium. What is the equilibrium wage for merely asking questions?
Of course academic life and projects will continue, but the real rewards will go to people doing new, innovative, and hitherto impossible projects with AI.
Big if true
Several important questions — such as the possibility of debt-rollover without primary surpluses — turn on whether the present value of the aggregate endowment is finite, i.e., whether the economic growth rate under the “risk-neutral” measure, lies below the risk-free rate. It is tempting to argue that the endowment must be finitely valued, since there exist finitely-valued, non-depreciating assets whose cash flows are cointegrated with aggregate output. This paper shows why this argument is incorrect. A remarkable historical episode in which French government bonds were indexed to aggregate growth allows direct measurement of the risk-adjusted growth rate, which is found to exceed the risk-free rate.
That is from a new NBER working paper by Stavros Panageas.
The US Exports Intelligence
Most Americans work in the service sector so it’s not surprising that most export-related jobs are in the service sector (The U.S. exports about $2.2 trillion of goods and $1.2 trillion of services, but services are more labor intensive than manufacturing so they support more export jobs per dollar.)
Richard Baldwin writes:
In 2022, US service exports supported 8.9 million American jobs.
US manufacturing exports supported 2.2 million.
That’s four-to-one in favour of services. Yet in the national narrative, ‘export jobs’ almost always means things done in steel mills and factories.
…When a household in Germany pays for Netflix, that is an American export. When a Brazilian retailer buys Microsoft cloud capacity, that is an American export. When JPMorgan structures a financial deal in London, or an American consulting firm advises a company in Singapore, those are American exports too.
None of these is shipped in a container. No customs official records them as they clear the customshouse. Yet they are exports since they earn foreign income for America just as surely as the ‘Boeings, Beans and Beef’ that President Trump sold on his recent China trip.
Need I remind you that when OpenAI sells intelligence to people abroad, that is a US export? N.B. this is the future.
World trade in goods expanded roughly five-fold between 1990 and 2020. Trade in digitally enabled services expanded more than eleven-fold over the same period. These are the modern services.
The trade debate is fixated on manufacturing—where America is doing fine—while largely ignoring services, where America is crushing. Increasingly, our most valuable exports travel not on container ships but at the speed of light over fiber.