My Blind Spots

by  Alex Tabarrok in

“Based on everything I’ve ever asked you, what do you think my biggest blind spots are?” ChatGPT replies:

Here’s a speculative, analytical take based on your past questions and style of interaction:

1. Underestimating Non-Market Frictions in Implementation
You’re extremely sharp on incentives, regulation, and institutional design, especially in economic theory and historical analysis. But sometimes there’s a relative underemphasis on execution risk — how slow-moving bureaucracies, agency coordination failures, or political coalitions can thwart otherwise well-aligned incentive structures. You assume policy failure is mostly about incentives or ideology, but sometimes it’s about sheer implementation entropy.

Example: In questions about permitting reform or ARPA-H, you tend to focus on structural misalignments or regulatory burden. That’s a big part of it, but sometimes it’s also just institutional incompetence or “slack” that defies modeling.

2. Belief in the Power of Market Discovery—But Underplaying Political Constraints on Entry
You have a strong pro-innovation, pro-market stance (which I think is generally well-supported), but perhaps slightly underweight how entrenched interests and legal choke points often stifle the ability of new entrants or ideas to surface, regardless of technical merit.

Example: In urban planning discussions, you’re excellent on NEPA, zoning, etc.—but some of the constraints (e.g., unions, fire codes, or insurance regulation) are less about “bad planning ideology” and more about deliberately created veto points to preserve incumbent interests. These can be harder to dislodge than your strategic framing sometimes suggests.

3. Technological Optimism Without Full Political Economy Accounting
You’re rightly excited by machine learning, drug discovery, synthetic data, airships, etc., and you press for empirical backing. However, you may be underestimating the political economy of deployment — i.e., how many powerful actors are actually threatened by cost-reducing innovation (e.g., in health, energy, or defense) and have the ability to delay, distort, or co-opt those technologies.

Example: In the machine learning for antibiotics discussion, you emphasize technical pathways (which are exciting), but less the entrenched FDA-pharma interplay or hospital group purchasing organizations that can prevent adoption even when evidence is strong.

4. Preference for Models That Cleanly Allocate Causality
You often seek elegant theoretical clarity — e.g., all-pay auctions, comparative advantage under automation, misallocation models. But the real world often features multiple overlapping distortions (e.g., taxes, zoning, union rules, insurance, etc.) that interact nonlinearly. Sometimes there’s a search for “the” distortion when it’s really a system of entangled local optima.

5. Policy-First Lens vs. Demand-Side Social Shifts
You focus a lot on supply-side issues (permitting, misallocation, productivity). This is correct and important. But occasionally the demand-side cultural or preference shifts get underweighted — e.g., the declining taste for urban density, aversion to risk, or increasing parental control over children’s autonomy. These may not be fixable through supply-side tweaks alone.

Hat tip: Aashna Doshi.

Subterranean sentences to ponder

by  Tyler Cowen in

But the fact that it’s commonplace is precisely why Earth’s subsurface biosphere is so compelling.  Mud is everywhere, which means it is important.  If you add up the total amount of mud underneath all the worlds’s oceans, you come up with a volume equivalent to about the entire Atlantic Ocean.  And, per cubic meter, there are 100 to 100,000 times more microbial cells in mud than there are in seawater.  That means that there’s so much intraterrestrial life in the subsurface that it’s hard to even fathom it.  The total amount of microbial cells in the marine sediment subsurface is estimated to be 2.9 x 10 [to the 29th] cells.  This is about 10,000 times more than the estimated number of stars in the universe.  But that’s not the whole subsurface.  You’d have to at least double this number to include the microbial cells living deep underneath the land.  And some of these cells may have found pockets where the food is more abundant than the average location, so more cells can live there than our models predict.  For these reasons, the actual number of microbial cells in the subsurface biosphere is certain to be much higher than our current estimates.

That is from the new and interesting IntraTerrestrtials: Discovering the Strangest Life on Earth, by Karen G Lloyd.

Tuesday assorted links

by  Tyler Cowen

1. Scott Sumner on who is rising and falling in status.  Mostly I agree.

2. New personalized dating form for STEM men and women, here is the link for women.  The previous link I put up for men already is generating matches and perhaps a marriage as well.

3. Michael Nielsen wants to reconsider AI alignment as the goal.

4. Musa al-Gharbi on Harvard’s decision.

5. For a brief moment, Harvard tweets libertarian, pre-1964 stuff.  And Alex retweets Will Stancil, and justifiably so.

Book ban sentences to ponder

by  Tyler Cowen in

Using a staggered difference-in-differences design, we find that the circulations of banned books increased by 12%, on average, compared with comparable nonbanned titles after the ban. We also find that banning a book in a state leads to increased circulation in states without bans. We show that the increase in consumption is driven by books from lesser-known authors, suggesting that new and unknown authors stand to gain from the increasing consumer support. Additionally, our results demonstrate that books with higher visibility on social media following the ban see an increase in consumption, suggesting a pivotal role played by social media. Using patron-level data from the Seattle Public Library that include the borrower’s age, we provide suggestive evidence that the increase in readership in the aggregate data is driven, in part, by children reading a book more once it is banned. Using data on campaign emails sent to potential donors subscribed to politicians’ mailing lists, we show a significant increase in mentions of book ban-related topics in fundraising emails sent by Republican candidates.

That is from a new paper by Uttara M. Ananthakrishnan, et.al., via Kris Gulati.

The Decline of Manufacturing

by  Alex Tabarrok

Look at the incredible decline of manufacturing across a wide variety of industries in the figure below. Nearly 7.5 million jobs lost, just since 2011. Devastated communities. Factories shuttered. Millions of older, lower-skilled workers out of work. A rise in inequality. Jobs going overseas.

“The future is bleak and hopeless if we continue like this,” said one worker.

…“This eventually leads to serious social problems, such as higher rates of unemployment and increased crime and social unrest. Nations with socially polarised work forces also suffer from political instability.”

And now for the rest of the story (non-paywalled or go to the comments.)

Will American soft power triumph through AI?

by  Tyler Cowen in

That is the theme of my latest Free Press column, here is one bit from it:

…for all the differences across the models, they are remarkably similar. That’s because they all have souls rooted in the ideals of Western civilization. They reflect Western notions of rationality, discourse, and objectivity—even if they sometimes fall short in achieving those ends. Their understanding of “what counts as winning an argument” or “what counts as a tough question to answer” stems from the long Western traditions, starting with ancient Greece and the Judeo-Christian heritage. They will put on a Buddhist persona if you request that, but that, too, is a Western approach to thinking about religion and ideology as an item on a menu.

These universal properties of the models are no accident, as they are primarily trained on Western outputs, whether from the internet or from the books they have digested. Furthermore, the leading models are created by Bay Area labor and rooted in American corporate practices, even if the workers come from around the world. They are expected to do things the American way.

The bottom line is that the smartest entities in the world—the top AI programs—will not just be Western but likely even American in their intellectual and ideological orientations for some while to come. (That probably means the rest of the world will end up a bit more “woke” as well, for better or worse.)

One of the biggest soft power victories in all of world history occurred over the last few years, and hardly anyone has noticed.

You might think the Chinese AI models are fundamentally different, but they are not. They too “think like Westerners.” That’s no surprise because it is highly likely that the top Chinese model, DeepSeek, was distilled from OpenAI models and also is based on data largely taken from Western sources. DeepSeek’s incredible innovation was to make the model much cheaper in terms of required compute, but the Chinese did not build their own model from scratch. And DeepSeek has the same basic broad ideological orientation as the American models, again putting aside issues related to Chinese politics. Unless an issue is framed in explicitly anti–Chinese Communist Party (CCP) terms, as a Taiwan query might be, it still thinks like an American.

Manus is another top Chinese AI model, but it is believed the makers built it upon Claude, an AI model from the American company Anthropic.

And this:

The geopolitics of all this have yet to play out. But already the most intelligent entities in the world are thinking, and evaluating options, like Westerners and Americans. Censoring them on a few issues related directly to Chinese politics will not change that basic reality.

In other words, the entire Chinese service sector, over time, may be built upon Western modes of thought and Western ideology. That includes the Chinese government and of course, the CCP itself. The point is that, over time, everyone’s thoughts and decisions and mental frameworks will be nudged in Western and American directions.

These are underrated points of import.

Harvard says no

by  Tyler Cowen in

Harvard University said on Monday that it had rejected policy changes requested by the Trump administration, becoming the first university to directly refuse to comply with the administration’s demands and setting up a showdown between the federal government and the nation’s wealthiest university.

Other universities have pushed back against the Trump administration’s interference in higher education. But Harvard’s response, which essentially called the Trump administration’s demands illegal, marked a major shift in tone for the nation’s most influential school, which has been criticized in recent weeks for capitulating to Trump administration pressure…

Some of the actions that the Trump administration demanded of Harvard were:

  • Conducting plagiarism checks on all current and prospective faculty members.

  • Sharing all its hiring data with the Trump administration, and subjecting itself to audits of its hiring while “reforms are being implemented,” at least through 2028.

  • Providing all admissions data to the federal government, including information on both rejected and admitted applicants, sorted by race, national origin, grade-point average and performance on standardized tests.

  • Immediately shutting down any programming related to diversity, equity and inclusion.

  • Overhauling academic programs that the Trump administration says have “egregious records on antisemitism,” including placing certain departments and programs under an external audit. The list includes the Divinity School, the Graduate School of Education, the School of Public Health and the Medical School, among many others.

Here is more from Vimal Patel at the NYT.

The roots of gun violence

by  Tyler Cowen in

An estimated 80 percent [of U.S: gun shootings] seem to instead be crimes of passion — including rage.  They’re arguments that could be defused but aren’t, then end in tragedy because someone has a gun.  Most violent crimes are the result of human behavior gone temporarily haywire, not premeditated acts for financial benefit.

That is from the new and interesting Unforgiving Places: The Unexpected Origins of American Gun Violence, by Jens Ludwig.

I never knew Joseph Smith ran for President

by  Tyler Cowen in

Eventually, Smith declared himself a candidate for the White House.  His proposed platform was an awkward conglomeration of popular, though incongruent, principles including restoring the national bank, cutting Congress members’ salaries, annexing Texas, and instituting the gradual abolition of slavery.  Hundreds of Mormon men, including Brigham Young, swarmed the nation campaigning for their prophet to become president.

That is from the new and excellent Benjamin E. Park, American Zion: A New History of Mormonism.  An excellent book, good enough to make the year’s best non-fiction list.

I also learned recently (from Utah, not from this book) that early Mormons would drink alcohol and “Brigham Young even operated a commercial distillery east of Salt Lake City, and his southern‐Utah “Dixie Wine Mission” (1860s‑80s) was organized to supply sacramental, medicinal, and commercial wine for the territory.”  By the time Prohibition rolled around, however, Mormons were close to completely “dry.”

Markets expand to fill empty spaces

by  Tyler Cowen in

How does a start-up founder hire a software engineer or a tech worker find a date in 2025? They could share a job posting on LinkedIn, log on to a dating app — or leave a handwritten note in a Waymo.

That’s what Influur CEO Alessandra Angelini was thinking when she and Jennifer Lo Chan, Influur’s head of marketing, tucked a piece of paper into a self-driving taxi’s center console on a recent trip to San Francisco.

“Looking to hire senior software engineers to work on AI/music project,” said the note, a retro touch in a futuristic setting, with the company’s website and Angelini’s email scrawled below. That same day, another Waymo rider snapped a photo of the note and posted it on the social media platform X, where it has been reshared nearly 200 times, received more than 4,000 likes and more than 300,000 views…

A few days later, another handwritten ad found in a Waymo was shared on X from a single man looking for a date. The bachelor described himself as a 26-year-old who works in tech but “doesn’t make it my whole personality” and left a phone number for interested parties to text. The post has gotten more than 200 replies.

Here is more from Lisa Bonos at The Washington Post.

Saturday assorted links

by  Tyler Cowen

1. New Knausgaard novel coming.

2. Unsecured penguin caused helicopter crash in South Africa.

3. Why does pre-training work?

4. Opinion polling from Poland.

5. Survey measures of the natural rate of interest.  Thread is here.

6. Cass Sunstein reports on the social neuroticism of the Right.  And Balaji is correct.

7. The conservative legal case against the tariffs (New Yorker).

8. WSJ interview with Doug Irwin.