A new economic model of AI and automation
Here is but one part of the results:
Given complementarity between the two sectors, the marginal returns to intelligence saturate, no matter how fast AI scales. Because the price of AI capital is falling much faster than that of physical capital, intelligence tasks are automated first, pushing human labor toward the physical sector. The impact of automation on wages is theoretically ambiguous and can be non-monotonic in the degree of automation. A necessary condition for automation to decrease wages is that the share of employment in the intelligence sector decreases; this condition is not sufficient because automation can raise output enough to offset negative reallocation effects. In our baseline simulation, wages increase and then decrease with automation.
That is from Konrad Kording and Ioana Elena Marinescu of the University of Pennsylvania. I am very glad to see ongoing progress in this area. Via the excellent Kevin Lewis.
Tuesday assorted links
1. Pre-history of progress studies and links.
2. Do institutional investors raise housing prices?
3. Alex Sarr leads the NBA in blocks.
4. History LLMs.
5. The new Middle Eastern Cold War. Likely to be one of the better and most important of essays from this year.
My Win-Win podcast with Liv Boeree
Liv is great at this, here is the Spotify link. Note this was recorded in May 2025, and its release postponed due to technical difficulties. So if a few parts seem “behind the times,” that is why. ” Tyler also shares his views on economic growth, UBI, automation, persuasion, state capacity, why fears of mass unemployment and civilizational collapse are often overstated.”
Grade inflation sentences to ponder
Next, we consider the effects of grade inflation on future outcomes. Passing grade inflation reduces the likelihood of being held back, increases high school graduation, and increases initial enrollment in two-year colleges. Mean grade inflation reduces future test scores, reduces the likelihood of graduating from high school, reduces college enrollment, and ultimately reduces earnings.
Here is the full paper by Jeffrey T. Denning, Rachel Nesbit, Nolan Pope, and Merrill Warnick. Via Kris Gulati.
Claims about AI productivity improvements
This paper derives “Scaling Laws for Economic Impacts”- empirical relationships between the training compute of Large Language Models (LLMs) and professional productivity. In a preregistered experiment, over 500 consultants, data analysts, and managers completed professional tasks using one of 13 LLMs. We find that each year of model progress reduced task time by 8%, with 56% of gains driven by increased compute and 44% by algorithmic progress. However, productivity gains were significantly larger for non-agentic analytical tasks compared to agentic workflows requiring tool use. These findings suggest continued model scaling could boost U.S. productivity by approximately 20% over the next decade.
That is from Ali Merali of Yale University.
Venezuela stablecoin fact of the day
By one estimate, almost 80% of Venezuela’s oil revenue is collected in stablecoins like tether, a local economist, Asdrúbal Oliveros, said on a recent podcast.
Here is the full WSJ piece.
The downside of NAFTA?
We study how NAFTA changed the geography of violence in Mexico. We propose that this open border policy increased trafficking profits of Mexican cartels, resulting in violent competition among them. We test this hypothesis by comparing changes in drug-related homicides after NAFTA’s introduction in 1994 across municipalities with and without drug-trafficking routes. Routes are predicted least cost paths connecting municipalities with a recent history of detected drug trafficking with U.S. land ports of entry. On these routes, homicides increase by 2.1 per 100,000 inhabitants, which is equivalent to 26% of the pre-NAFTA mean. These results cannot be explained by changes in worker’s opportunity costs of using violence resulting from the trade shock.
That is from a new JDE paper by Eduardo Hidalgo, Erik Horning, and Pablo Selaya. Via the excellent Kevin Lewis.
Monday assorted links
1. The (strong) case for beans (WSJ).
2. Education is correlated with liberal and pro-market views in most countries.
3. Puffin photos.
4. Do GLP-1 drugs pay for themselves?
6. New Statesman recommends non-fiction for 2026.
7. David Deming on learning with generative AI.
9. Weird LLM generalizations. With a good Terminator example.
What should I ask Joe Studwell?
He has a new and excellent book coming out, namely How Africa Works: Success and Failure on the World’s Last Developmental Frontier, which I consumed eagerly. You probably know his earlier book How Asia Works. So what should I ask him?
For additional context, here is the opening of his home page (no Wikipedia page?):
Hello. I am an author, journalist, public speaker and occasional university teacher. I am based much of the time in Cambridge. In the 2000s I restored and lived in a home in a still unspoiled area of central Italy (the photo at the top of the page is a view from the house).
So what should I ask him?
MRU college fellowship
The MRU College Fellowship, for US undergrad and grad students, helps Fellows produce their own videos, podcasts, or other online content to bring economic insights to a wide audience. Fellows are paired with MRU mentors for a seven-week remote program, starting with an expenses-paid weekend kickoff event in DC. Fellows also earn a $2,000 stipend.
Applications are due January 23.
Low-skilled immigration into the UK
I asked GPT 5.2 Pro what it thought of the welfare consequences of UK immigration, and here are its summary remarks:
The literature does not support the claim that low-skilled immigration has imposed large net welfare losses on the UK as a whole. Instead, it supports something like:
-
Net welfare for existing residents is likely modestly positive (or near zero but not strongly negative) on average,
-
but the distributional impacts can be meaningfully negative for some low-skilled native workers and for some localities,
-
and the sign/magnitude hinge heavily on productivity spillovers and on dynamic trajectories (skill acquisition, occupational mobility, family formation).
The entire response is useful and well thought out.
Sunday assorted links
1. Request for research proposals on psychology of progress.
2. Interview on NIH grants and how to improve them.
3. Burry, Jack Clark, and D. Patel.
4. a16z. A good and impressive piece.
5. That California tax would hit some people pretty hard.
6. NYT on the history of picture books.
My Austin visit
First, I gave a talk at University of Austin and also had some meetings there, including with students. My talk was a practical guide on how to use AI to offer courses that a college or university otherwise cannot afford (especially important for smaller institutions). I believe they will be putting it online.
My general sense was that U. Austin undergraduates are on a par with undergraduates at top five schools. I do not think on the technical side they would compete with Stanford or MIT, but more generally…they were very impressive and asked excellent questions with real curiosity. And seemed politically saner than typical Ivy League cohorts, though without being “mono” in any particular direction. Here is Arnold Kling on UATX and its students.
The school does admissions by SAT scores only.
Austin is also one of my favorite places to eat in the United States. It is especially strong in areas of import to me, including barbecue, cheeseburgers, and Tex-Mex. Just ask your local friendly LLM…
Those new service sector jobs
Basketball Expert (Fans, Journalist, Commentator, etc.)
Role Overview
We’re looking for Basketball experts — avid fans, sports journalists, commentators, and former or semi-professional players — to evaluate basketball games. You’ll watch basketball games and answer questions in real time assessing the quality, depth, and accuracy of AI insights, helping us refine our AI’s basketball reasoning, storytelling, and strategic understanding.
Key Responsibilities
Game Evaluation: Watch basketball games and review AI-generated play-by-play commentary and post-game analysis.
Performance Scoring: Rate the accuracy, insight, and entertainment value of AI sports coverage.
Context & Understanding: Assess the AI’s grasp of player performance, game flow, and strategic decisions.
Error Detection: Identify factual mistakes, poor interpretations, or stylistic inconsistencies.
Feedback Reporting: Provide clear written feedback highlighting strengths, weaknesses, and improvement opportunities.
Collaboration: Work with analysts and developers to enhance the AI’s basketball-specific reasoning and realism.
From Mercor, pays $45 to $70 an hour. For background on Mercor, see my very recent CWT with Brendan Foody. Via Mike Rosenwald, wonderful NYT obituary from him here.
Congress is reversing Trump’s budget cuts to science
Surprisingly, analysts foresee a possible rise of more than 2 percent in the budget category known as basic research — the blue-sky variety that produces fundamental strides and spinoffs in fields such as health care and artificial intelligence. Last year, the Trump administration called for a cut in federal basic research of more than one-third.
Mr. Trump sought even larger cuts for the National Science Foundation, which sponsors much of the nation’s basic research. He proposed that its budget be slashed to $3.9 billion from $8.8 billion, a drop of 56 percent. The Senate package countered with a reduction to $8.75 billion, or less than 1 percent.
The bipartisan accord on funding science, Ms. Zimmermann said, stands in sharp contrast with the congressional impasse that shut down the government last fall as Democrats and Republicans clashed over the renewal of subsidies for the Affordable Care Act.
“They’re working together now,” she said. “It’s a return to normalcy.” The new cooperation, Ms. Zimmermann added, is “promising for the eventual passage of the bills.”
Here is the full NYT article.