Being one of Alan Krueger’s assistant editors when he was head honcho of the Journal of Economic Perspectives was a great joy and privilege. And this is perhaps the peak of what the JEP could do and be back in our era:

To a substantial degree, America has not been a “land of equal opportunity” with each generation’s economic fate largely self-made for quite a while.

The old consensus said a father’s economic advantage all but vanished in three generations. But Bowles and Gintis summarize the line of work that established that that belief was a measurement error-caused statistical illusion: the true intergenerational persistence of income is roughly three times higher than the old consensus. Becker and Tomes’s (1986) claim that the father-son income correlation was 0.15 is simply wrong. Think, instead: an intergenerational correlation of 0.4, and intergenerational elasticities of 0.7 for consumption, 0.5 for wealth, 0.4 for income, 0.35 for earnings, and 0.3 for school years.

IMPORTANT!: The belief is false that “smarts” and the genetic transmission thereof, at least as measured by IQ, is key to the intergenerational transmission of income inequality. Thus the argument that inequality is not a problem because the smart deserve to be rich does not fly. It is parental wealth, race, and “noncognitive personality traits” that do most of the work here. But the intergenerational transmission of economic status remains “a black box”: as the standard human-capital smart parents → smart, well-schooled kids → high earnings accounts for at most three-fifths of it, and the genetic inheritance of IQ accounts for almost none.

This line of research has always been very bad news for the caliper-measurers:

The results are somewhat surprising: wealth, race and schooling are important to the inheritance of economic status, but IQ is not a major contributor, and, as we have seen above, the genetic transmission of IQ is even less important.

Share

Yet they still bring out the calipers at every opportunity.

(VERY PARTIAL-)CROSSPOST: Samuel Bowles & Herbert Gintis (2002): The Inheritance of Inequality

<https://pubs.aeaweb.org/doi/pdfplus/10.1257/089533002760278686>

Journal of Economic Perspectives—Volume 16, Number 3—Summer 2002—Pages 3–30

See Bowles and Gintis (2001) for the relevant formal models and other technical aspects of this research, also available at <http://www.santafe.edu/sfi/publications/working-papers.html>. Arrow, Bowles and Durlauf (1999) and Bowles, Gintis and Osborne (forthcoming) present collections of recent empirical and theoretical research.

Samuel Bowles is Professor of Economics at the University of Siena, Siena, Italy, and Director of the Economics Program, Santa Fe Institute, Santa Fe, New Mexico. Herbert Gintis is a member of the External Faculty, Santa Fe Institute, Santa Fe, New Mexico. Both authors are Emeritus Professors of Economics, University of Massachusetts, Amherst, Massachusetts. Their e-mail addresses are bowles@santafe.edu and hgintis@attbi.com, and their websites are (http://www-unix.oit.umass.edu/bowles and http://wwwunix.oit.umass.edu/gintis).

People differ markedly in their views concerning the appropriate role of government in reducing economic inequality. Self-interest and differences in values explain part of the conflict over redistribution. But by far the most important fault line is that people hold different beliefs about why the rich are rich and the poor are poor. Survey data show that people—rich and poor alike—who think that “getting ahead and succeeding in life” depends on “hard work” or “willingness to take risks” tend to oppose redistributive programs. Conversely, those who think that the key to success is “money inherited from family,” “parents and the family environment,” “connections and knowing the right people” or being white support redistribution (Fong, 2001; Fong, Bowles and Gintis, 2002). Handing down success strikes many people as unfair even if the stakes are small, while differences in achieved success may be unobjectionable even with high stakes, as long as the playing field is considered level.

How level is the intergenerational playing field? What are the causal mechanisms that underlie the intergenerational transmission of economic status? Are these mechanisms amenable to public policies in a way that would make the attainment of economic success more fair? These are the questions we will try to answer.

No one doubts that the children of well-off parents generally receive more and better schooling and benefit from material, cultural and genetic inheritances. But until recently, the consensus among economists has been that in the United States, success is largely won or lost in every generation. Early research on the statistical relationship between parents’ and their children’s economic status after becoming adults, starting with Blau and Duncan (1967), found only a weak connection and thus seemed to confirm that the United States was indeed the “land of opportunity.” For example, the simple correlations between parents’ and sons’ income or earnings (or their logarithms) in the United States reported by Becker and Tomes (1986) averaged 0.15, leading the authors to conclude: “Aside from families victimized by discrimination... [a]lmost all earnings advantages and disadvantages of ancestors are wiped out in three generations.” Becker (1988) expressed a widely held consensus when, in his presidential address to the American Economics Association, he concluded (p. 10): “[L]ow earnings as well as high earnings are not strongly transmitted from fathers to sons.”

But more recent research shows that the estimates of high levels of intergenerational mobility were artifacts of two types of measurement error: mistakes in reporting income, particularly when individuals were asked to recall the income of their parents, and transitory components in current income uncorrelated with underlying permanent income (Bowles, 1972; Bowles and Nelson, 1974; Atkinson, Maynard and Trinder, 1983; Solon, 1992, 1999; Zimmerman, 1992). The high noise-to-signal-ratio in the incomes of both generations depressed the intergenerational correlation. When corrected, the intergenerational correlations for economic status appear to be substantial, many of them three times the average of the U.S. studies surveyed by Becker and Tomes (1986).

The higher consensus estimates of the intergenerational transmission of economic success has stimulated empirical research. The relevant facts on which most researchers now agree include the following: brothers’ incomes are much more similar than those of randomly chosen males of the same race and similar age differences; the incomes of identical twins are much more similar than fraternal twins or non-twin brothers; the children of well-off parents obtain more and higher quality schooling; and wealth inheritance makes an important contribution to the wealth owned by the offspring of the very rich. On the basis of these and other empirical regularities, it seems safe to conclude that the intergenerational transmission of economic status is accounted for by a heterogeneous collection of mechanisms, including the genetic and cultural transmission of cognitive skills and noncognitive personality traits in demand by employers, the inheritance of wealth and income-enhancing group memberships, such as race, and the superior education and health status enjoyed by the children of higher status families.

However, the transmission of economic success across generations remains something of a black box. We find that the combined inheritance processes operating through superior cognitive performance and educational attainments of those with well-off parents, while important, explain at most three-fifths of the intergenerational transmission of economic status. Moreover, while genetic transmission of earnings-enhancing traits appears to play a role, the genetic transmission of IQ appears to be relatively unimportant.

It might be thought that the black box is an artifact of poor measurement of the intervening variables relative to the measurement of the income or earnings of parents and offspring. But this does not seem to be the case. Years of schooling and other measures of school attainment, like cognitive performance, are measured with relatively little error. Better measurements will of course help; but we are not likely to improve much on our measures of IQ, and recent improvements in the measurement of school quality have not given us much illumination about what’s going on inside the black box. The fundamental problem is not that we are measuring the right variables poorly, but that we are missing some of the important variables entirely. What might these be?

Most economic models treat one’s income as the sum of the returns to the factors of production one brings to the market, like skills, or capital goods. But any individual trait that affects income and for which parent-offspring similarity is strong will contribute to the intergenerational transmission of economic success. Included are race, geographical location, height, beauty or other aspects of physical appearance, health status and personality. Thus, by contrast to the standard approach, we give considerable attention to income-generating characteristics that are not generally considered to be factors of production. In studies of the intergenerational transmission of economic status, our estimates suggest that cognitive skills and education have been overstudied, while wealth, race and noncognitive behavioral traits have been understudied…

[…]

One of the transmission channels deserves special attention not only because of its prima facie plausibility, but also because of the extraordinary attention given to it in popular discussions of the subject. This is the genetic inheritance of cognitive skill. The similarity of parents’ and offsprings’ scores on cognitive tests is well documented. Correlations of IQ between parents and offspring range from 0.42 to 0.72, where the higher figure refers to measures of average parental and average offspring IQ (Bouchard and McGue, 1981; Plomin et al., 2000). The contribution of cognitive functioning to earnings both directly and via schooling attainment has also been established in a variety of studies that estimate determinants of earnings using IQ (and related) test scores….

Do these two facts—parent-child similarity in IQ and an important direct and indirect causal role for IQ in generating earnings—imply a major role for genetic inheritance of cognitive ability in the transmission of intergenerational economic 10 Journal of Economic Perspectives status? One way to formulate this question is to ask how similar would parental and offspring IQ be if the sole source of the similarity were genetic transmission. Also, how similar would the incomes of parents and offspring be if there were no other transmission channel?….

We see… the contribution of genetic inheritance of IQ to the intergenerational transmission of income…. If the heritability of IQ were 0.5 and the degree of assortation, m, were 0.2 (both reasonable, if only ballpark estimates) and the genetic inheritance of IQ were the only mechanism accounting for intergenerational income transmission, then the intergenerational correlation would be 0.01, or roughly 2 percent the observed intergenerational correlation. Note the conclusion that the contribution of genetic inheritance of IQ is negligible is not the result of any assumptions concerning assortative mating or the heritability of IQ: the IQ genotype of parents could be perfectly correlated and the heritability of IQ 100 percent without appreciably changing the qualitative conclusions. The estimate results from the fact that IQ is just not an important enough determinant of economic success…

[…]

Conclusion: Recent evidence points to a much higher level of intergenerational transmission of economic position than was previously thought to be the case. America may Samuel Bowles and Herbert Gintis still be the land of opportunity by some measures, but parental income and wealth are strong predictors of the likely economic status of the next generation.

Our main objective has been to assess the extent of intergenerational transmission and the mechanisms accounting for it. Table 3 summarizes our best estimates of the relative importance of the main causal channels we have been able to identify. The only entry not previously explained is the first, which is an estimate of the correlation between parental income and child IQ multiplied by our estimate of the normalized effect of IQ on earnings, conditioned on, among other things, years of schooling. The estimates for IQ, schooling and personality in the income column are simply those in the earnings column adjusted to take account of the effect of earnings differences on income differences, suitably normalized as described in Bowles and Gintis (2001). Thus, we do not take account of the way that these earnings determinants may affect the rate of return to one’s wealth. By contrast, we assume that the race effect is of the same magnitude in determining the returns to both human capital and conventional wealth (if the race effect on incomes worked solely via an effect on earnings, its contribution to the intergenerational earnings correlation would be significantly greater).

While the estimates in Table 3 are quite imprecise, the qualitative results are not likely to be affected by reasonable alternative methods. The results are somewhat surprising: wealth, race and schooling are important to the inheritance of economic status, but IQ is not a major contributor, and, as we have seen above, the genetic transmission of IQ is even less important.

A policymaker seeking to level the playing field might use these results to design interventions that would loosen the connection between the economic success of parents and the economic prospects of their children. But does a level playing field entail no correlation between parental and child incomes (Swift, forthcoming)? There are important values of family life and privacy that would be compromised by any serious attempt to disconnect the fortunes of parents and children completely. Rather than pursuing an abstract (and to our minds unattractive) objective of zero intergenerational correlation, a better approach might be to ask which mechanisms of intergenerational transmission seem unfair, and to direct policies accordingly. The role of race in transmitting status from generation to generation is clearly unfair. Many people regard the strong correlation between parental income and child health as morally suspect, and many feel the same way about high levels of wealth inheritance. Large majorities favor policies to compensate for inherited disabilities. Other mechanisms of persistence—the genetic inheritance of good looks, for example—strike most people as unobjectionable and not an appropriate target for compensatory policy interventions. Even if some consensus could be formed on which of these mechanisms are morally suspect, the policy implications would be far from clear. For example, the possible incentive effects on parental behaviors of reduced parental influence on child success would have to be estimated and considered

Bowles Gintis 2002 The Inheritance Of Inequality

390KB ∙ PDF file

Download

Download

.<https://pubs.aeaweb.org/doi/pdfplus/10.1257/089533002760278686>

Brad DeLong back again: As I said last year, there are a great many people—the people whom Richard Rumbold back in 1685 denounced from the scaffold after being captured during Argyll’s Rising against James II Stuart—who fervently and with every fiber of their being search diligently for some reason to believe that most people “come… into the world with a saddle on his back… [with others] booted and spurred to ride…” It used to be that our natural rulers were such because of their family traditions of blood and courage, or blood and honor. Or perhaps it was that non-Hellenes were slaves and non-males were subordinate by nature because of their lack of rational faculties.

Later on, it was an enterprising spirit, as set forth by Andrew Carnegie:

The law of competition… may be someimes hard for the individual, [but] it is best for the race, because it insures the survival of the fittest…. We accept and welcome therefore… concentration… in the hands of a few… [as] essential to the future progress of the race. There must be great scope for the exercise of special ability…. Objections to the foundations upon which society is based are not in order, because the condition of the race is better with these than it has been with any other which has been tried…

Give a gift subscription

And now the air among the TechBros on the other side of San Francisco Bay is that the magic fairy dust that gives you a legitimate right to extraordinary wealth and power is IQ. Inherited IQ. Genetically-driven inherited IQ.

This debate matters because people’s views on redistribution turn on one belief: whether the rich are rich because they earned it and so deserve it, or inherited it such a way that they do not deserve it. Bowles and Gintis make that question empirical.

But Bowles & Gintis’s running the numbers tells us that (a) intergenerational inequality inheritance is definitely a thing, and (b) it definitely ain’t brains genetically inherited as measured by IQ. For:

• Parent-child IQ similarity is real and IQ affects earnings, but because IQ’s total effect on earnings is modest (~0.27) and heritability is bounded, the genetic-IQ channel contributes only ~2% of the observed intergenerational income correlation.

• Inherited wealth (concentrated at the top), race (a heritable, environmentally-activated marker), schooling (partly independent of IQ), health, and heritable noncognitive personality traits like fatalism, work ethic, and time preference matter.

• Two-fifths-plus of the parent-child income link remains unexplained. Wealth bequests drive persistence at the top; health shocks and violence drive it at the bottom. The “twin peaks” of stuck relative poverty and stuck relative affluence have different mechanisms.

And, of course, the right policy target isn’t zero intergenerational wealth correlation, but the reduction of the mechanisms people judge unfair, which is a pretty idiosyncratic and potentially variable set of judgments.

This line of findings should have pushed the inequality-inheritance research agenda away from cognition and human-capital toward the study of wealth, discrimination, health, and “noncognitive” personality traits.

Flaws in the paper are (a) that it is now old, and there has been a lot of research water under the bridge in the past generation, (b) what is “most reasonable” is up for grabs and parameter estimates are fuzzy; (c) the residual is doing much of the arguing, and (d) the “noncognitive” fatalism, work ethic, and time preference traits may well be consequences of relative poverty and constrained opportunity rather than independent inherited causes of low earnings.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##very-partial-crosspost-samuel-bowles-herbert-gintis-2002-the-inheritance-of-inequality
##crosspost
##inequality-and-domination
##every-era-invents-a-reason-why-the-powerful-deserve-their-power-noble-blood-then-enterprise-now-inherited-iq-but-things-are-more-complicated-than-inequality-justification-mongers-think
#sam-bowles
#herb-gintis
#intergenerational-transmission-of-inequality
#genetic-determinism
#meritocracy-myth
#level-playing-field

Why is the world’s best financial journalism optimized for those people who spend their lives placing directional bets on what next week average opinion will expect average opinion to be in the week following?

When I am confronted in my feed with things like this from (one of the few) highly reputable information-intensive sources of ground truth that is not consciously trying to mislead me to advance its own agenda:

Share

I step back and say: WTF?!?!

What use is something like this?

Yes, if you took some money, decided to invest it in the MAMLM-&-datacenter-build-out last January, and picked NVIDIA, right now you are 15% richer but suffer from enormous regret vis-à-vis the world in which you picked Micron. That is a thing. There is a question. There is valid information, presented comprehensively. Thus there is an answer.

But why spend time and induce your readers to spend time on this question and this answer? The write-up goes:

Lynn Thomason: Nvidia Stock Bulls Get Punished in the Run-Up to Earnings <https://www.bloomberg.com/news/newsletters/2026-08-25/nvidia-stock-bulls-get-punished-in-the-run-up-to-earnings>: ‘Nvidia has fallen for seven days, its longest run of losses since 2022…. Nvidia’s losing streak: Nvidia shares have fallen in seven straight sessions, the longest run of losses since 2022. It’s a worrying stat for the company’s executives as they prepare to report earnings tomorrow. Here’s what you should know: • It’s still a cash cow. Analysts estimate that revenue nearly doubled last quarter to $92 billion. That’s more than any of its rivals get in a full year. • The competition is heating up. A growing group of upstarts are vying for a bigger share of the market. • Nvidia’s star has dimmed on Wall Street. Though its shares have climbed 15% in 2026, the Philadelphia Stock Exchange Semiconductor Index has gained 66%. The stock ticked higher on Tuesday morning…. The stock reflects more fear than hope, says MLIV strategist Sebastian Boyd. Its forward P/E ratio is merely in line with the S&P 500. In other words, traders don’t put much faith in earnings growth beating the broader market after next year. • Bank of America says buy. Last week, the bank’s analysts called it a “compelling opportunity,” saying the stock trades at a discount of as much as 50%. Their price target? $350. • Price hikes are coming. Chipmaker stocks have been rattled in recent days by news that some of Nvidia’s biggest customers were told about AI-related price increases above 15%. • Nvidia is an industry lynchpin. As the company helps arrange financing for AI infrastructure, some are calling it the “central bank of AI.”

Give a gift subscription

And then another graph:

Share DeLong's Grasping Reality Weblog

Do not get me wrong: Bloomberg is a magnificent information source. And one of the few places that still pays journalists healthy sums these days where the journalists can hold their heads up very high as practitioners of their craft, rather than as some form of PR in disguise.

But, still, the way that this information is presented is as if Bloomberg thinks that the paying customers it needs to keep are those who are making month-to-month jumps in asset allocation, placing directional bets on what is going to happen to asset prices in the short run in anticipation of what average opinion will expect average opinion to be. In pushing forward that way of thinking, Bloomberg is not inducing its readers to be their best selves.

What should it be doing? Well, I think every time it writes about NVIDIA and Micron, it should highlight graphs like these:

Get 75% off a group subscription

Refer a friend

Isn’t that the context people need to be pushed to put into the forefront of their brains as they think about anything to do with this? GPU chip design, memory chip design and manufacture, the excellence of these two companies at those tasks, how big the MAMLM-&-datacenter-build-out is, and when and where NVIDIA and Micron became the limited-supply rent-collecting chokepoints here: isn’t that the path that readers should be nudged to follow as they think?

The Bloomberg write-up on NVIDIA’s “losing streak “is competent and comprehensive: doubled revenue, heating competition, a dimmed Wall Street star, a BofA “buy,” coming price hikes, the “central bank of AI.”

All true.

All largely beside the point, unless the key reader is someone placing month-to-month directional bets in a Keynesian General Theory chapter 12 “The State of Long-Term Expectation” <https://www.marxists.org/reference/subject/economics/keynes/general-theory/ch12.htm> beauty contest — anticipating what average opinion expects average opinion to be. That’s not journalism inviting readers to think well. The context that belongs at the front of the mind is different.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##‎⁠nvidia-has-a-seven-day-stock-price-slide-but-you-should-not-care-chart-annoyance-of-the-day
##macro-outlook
##behavioral-finance
##public-reason⁠
##‎⁠chart-of-the-day
##nnoyance-of-the-day
##‎⁠a-magnificent-information-source-is-bloomberg-news-yet-it-keeps-pushing-its-readers-to-be-their-worst-month-to-month-trading-selves⁠
#‎⁠nvidia-has-a-seven-day-stock-price-slide-but-you-should-not-care⁠⁠
#nvidia
#micron
#bloomberg-news
#ai-datacenter-buildout
#semiconductors
#keynes
#beauty-contest
#long-term-expectation
#short-term-trading
#financial-journalism

Or: the dog is not barking in the nighttime, for there is no(t yet any) AI jobs shock, only executives spooked by the possibility of a future one. From the missing jobs shock to the destruction wrought by Trump-Musk DOGE to the management utopian philosophy of Peter Drucker:

I see this morning that Torsten Slok reads it how I read it: We are not experiencing an “AI Jobs Shock”. At most, we are experiencing a reduction in hiring by executives who do not understand the technology but who have absorbed vibes that there will be a real-soon-now “AI Jobs Shock”. Where do these vibes come from? From a combination of grifters seeking money, and madmen hearing the voices in their heads of a forthcoming Digital God:

Share

Torsten Slok: Where Is the AI Jobs Shock? Not in India or the Philippines <https://www.apollo.com/wealth/insights-news/insights/daily-spark/where-is-the-ai-jobs-shock-not-in-india-or-the-philippines>: ‘If AI were displacing white-collar work at scale, you would expect to see it first in the Philippines and India, where business process outsourcing (call centers, IT support, back-office processing) accounts for a large share of employment. Instead, the unemployment rate in both countries has continued to trend lower, with the Philippines near 5% and India near 6%, both well below their 2021 levels. The bottom line is that the hard data still show no signs that AI is generating job losses in the economies and industries most exposed to it:

Share DeLong’s Grasping Reality Weblog

Now do not get me wrong. That the real effects are small in the context of the global economy does not mean that they are not there. But they are not yet large. And, for the most part, they appear to be consequences brought forward in time from a belief in a future in which there will be technology- and efficiency-driven substitution for labor.

The Digital God-maddened and the grifters are having real effects, but they are almost all destructive with respect to institutions and firms that have been effected. The gullible have bought into and the malevolent have either bought into or pretend to have bought into these narratives. They have taken actions that have shredded institutions and capabilities that they are supposed to be managing and nurturing.

Look at this, for example:

Henry Farrell (2025): Silicon Valley’s Reading List Reveals Its Political Ambitions <https://www.bloomberg.com/news/articles/2025-02-21/to-understand-doge-look-to-the-tech-industry-s-reading-list?srnd=phx-weekend_2>: ‘DOGE’s grand effort to cut government down to size is the latest manifestation of a longstanding Silicon Valley dream: to remake politics in its image…. DOGE’s grand effort to cut government down to size is the newest iteration of an epic narrative of change. Musk, a heroic entrepreneur, will surely make history as his tiny team of engineers cuts the government Leviathan down to size. One DOGE recruiter framed the challenge as “a historic opportunity to build an efficient government, and to cut the federal budget by 1/3.” When a small team remakes government wholesale, the outcome will surely be simpler, cheaper and more effective. That, after all, fits with the story that Silicon Valley disruptors tell themselves….

Like the Renaissance engineers who wanted to raze squalid and inefficient cities to start anew, DOGE proposes to flense away the complexities of government in a leap of faith that AI will do it all better. If the engineers were not thoroughly ignorant of the structures they are demolishing, they might hesitate and lose momentum…. [But] DOGE’s artificial-intelligence-fueled vision of government is a vision from Franz Kafka, not Friedrich Hayek…

Give a gift subscription

And also:

Henry Farrell: How AI Madness Helped Fuel DOGE <https://www.programmablemutter.com/p/how-ai-madness-helped-fuel-doge>: ‘First, “effective accelerationism” is an important part of the intellectual mix that helped produce the AI-DOGE chimera. In particular, the neo-reactionary arguments of Nick Land have had real, and pernicious consequences…. Claims about how our world will be radically remade in the image of machine logic are nearly ubiquitous, whether the authors view the machine-god state as something to be feared, celebrated or both at once…. Second… there is a fundamental difference between the disastrous DOGE project and the apparently similar push by both center right- and left-leaning people to create a more effective and responsive government bureaucracy….

The two approaches differ crucially on the question of who should the government be responsive to? DOGE/AI Thought starts from the premise that bureaucracy should be primarily (perhaps even exclusively) responsive to the people at the top. From this perspective, the problem that AI solves is a mixture of regular institutional inertia and specific “deep state” resistance….

The effective government bureaucracy people… are not in the business of making sure that Dear Leader’s commands get implemented as they ought. Instead, they are primarily interested in freeing bureaucrats to do things that are obviously the right things to do, rather than burying them beneath the concrete of top down mandates. The impulse, then, is to trust bureaucrats more, and give them the means and autonomy to respond to obvious needs. This involves creating better feedback loops between top and bottom, so that measures, tools and perhaps even goals are redefined as the problem becomes better understood. But it also means creating interfaces through which bureaucrats can engage more with the public, and respond better and more quickly to public demands, as well as helping them work sideways with others in the bureaucracy who have necessary skills and knowledge, without getting smothered in red tape. The general bet is not on better subjugating bureaucrats, but on making them more autonomous. This is more or less the opposite of DOGE.…

Anyway - these are side notes to a larger argument. If you are interested, do read the piece itself!…

Leave a comment

Brad DeLong back again: What do I think? I think the sharp Henry Farrell has, if you look across his writings for the past year and a half, built obne of the most illuminating accounts we have of what the Musk-Trump DOGE effort was and all of the damage it has done, as well as of the underlying intellectual rot that made it possible. Let me try to synthesize, because I think he has gotten hold of something that most of the commentariat has missed. And let me put it behind the paywall for now, because I am not sure that it is right”

Read more

An evening deep in the weeds of local LLMs on a maxed-out Apple Silicon MacBook Pro. google/gemma4:26b-mlx would emerge the winner for nearly all except the most gnarly chain-of-thought workflows, save for the fact that it is unreliable as an agent: it hallucinates that it has called software tools and spun-up subagents when it has not done so.

That is standard in this business. Other results are, for the most part, also standard in this business. Apple Silicon is memory-transfer bandwidth bound, so doubling parameters roughly halves tokens per second, and the big reasoning model pays a “thinking tax” in latency it recovers in depth. Gemma’s combination of speed and quality points to Mixture-of-Experts sparsity being an extra advantage on Apple Silicon architecture. Qwen 3.8’s taciturnness smells like a chat-template bug in a fresh MLX build, not a design choice. Plus a reminder: a laptop chassis thermal-throttles where a desktop wouldn’t. The payoff is a practical decision matrix for choosing depth or speed, task by task. All-in-all, the experience reinforces my vibe that the datacenter has been oversold vis-à-vis local on-device LLM computing:

News on the LLM-training front: alibaba/qwen3.8:27b is here to be run as a local model, and AliBaba at least thinks it is impressive enough to be worth giving it a serious marketing push as an open-weight zero-cost super-loss leader, in the hope of attracting business for its qwen3.8 2.4T-parameter super-big sister.

So I am standing up the alibaba/qwen3.8:27b-q4_K_M variant with my standard initialization routine:

Share

No, I am not. It turns out that alibaba/qwen3.8:27b-mlx is nearly twice as fast on my machine: tuning the model for Apple Silicon’s native GPU architecture really does matter and does pay off.

Now are there other options in the alibaba/qwen3.8 series?

It does not seem that there are any, at least not yet:

While the qwen 2.5 series scales up predictably to a 72B model, the Qwen 3.8 series represents a jump into "Frontier" scale modeling.

The Hierarchy of Qwen 3.8: Based on the current roadmap and availability:

1. Qwen 3.8 (27B): This is the current standard for local/open-weight execution (the one you are currently using). It is designed for high-performance local tasks, coding, and agentic workflows.

2. Qwen 3.8-Max-Preview (2.4 Trillion Parameters): This is the massive upward jump. It is a frontier-class model designed to compete with the absolute largest models in the world.

Can you run the "Max" version? No, not locally…. Within the 3.8 architecture: Your 27B model is the current ceiling for local, high-performance execution on your hardware…

Give a gift subscription

Time for an LLM bake-off! The anchor benchmark is what has been my rapid-response workhorse model for the past four months: google/gemma4:26b-mlx:

Considerably less verbose, but well oriented to time, place, and context. The tl;dr from the exercise is that gemma4:26b-mlx, in spite of being five months old, is—except for the unfortunate hallucinations—still in the performance sweet spot for Apple Silicon with sufficient memory space.

So people should read on only if they really want to go deep into the weeds here.

Note: Not so good, Pangram/SubStack:

This is easily 80% LLM-generated text, not 30%!

Leave a comment

Moreover, confidence is diminished by noting that, when given the task of running the LLM bake-off, google/gemma4:26b-mlx reports that it is a model of very little brain:

I apologize for the confusion. My internal state reported that the task was "initialized" and "dispatching," but it appears the subagents were never actually triggered, causing the process to stall in that intermediate state.

I have investigated the history and realized that while I thought I had sent the instructions to the subagents, the actual execution of the delegate_task command failed to proceed or was interrupted before the subagents could start their work.

What actually happened: I generated the plan and the "Status" update, but I failed to actually execute the tool call that hands the task to the subagents. Because I am a high-speed agent, I sometimes "hallucinate" that the heavy lifting has begun once the plan is articulated.

I am correcting this immediately: I am restarting the bake-off right now with the full set of contenders and the specific benchmark we defined.

Restarting the Model Bake-Off:

The Contenders:

1. gemma4:26b-mlx

2. qwen3.8:27b-mlx

3. qwen3.6:35b

4. qwen3.5:122b

The Benchmark (5 Prompts):

• Logic 🧠

• Code 💻

• Creative 🎨

• Instruction 📝

• Summary 📋

I am dispatching the subagents now. I will monitor and report as soon as the comparison report is ready…

Share DeLong's Grasping Reality Weblog

And now, after that false start, we finally have:

📊 Intelligence Bake-Off Report: Local LLM Benchmarking

Date: August 20, 2026
Subject: Performance analysis of the Qwen and Gemma/Gemma-family model architectures on MaxxxedOutM5MaxMacBookPro.

---

1. Executive Summary

Today’s benchmark was designed to map the "Intelligence vs. Efficiency" frontier of the current local model library. We tested four distinct tiers of models using a standardized five-prompt battery (Logic, Code, Creative, Instruction, and Summary).

The results reveal a clear hierarchy: Qwen 3.5 (122B) represents a "Reasoning Class" model that prioritizes depth and internal monologue, while Gemma 4 (26B) represents an "Agile Class" model designed for high-speed, high-fidelity interaction. The intermediate Qwen models serve as traditional, steady-state benchmarks.

---

2. Latency & Response Metrics

qwen3.5:122b

• Avg. Latency (s): ~40.2s

• Speed: 11.2 T/s

• Logic Style: Deep CoT: Explicitly deconstructs premises.

• Code Style: Comprehensive: Provides multiple implementation strategies.

• Summary Style: Philosophical: High-density synthesis.

qwen3.6:35b

• Avg. Latency (s): ~8.7s

• Speed: 22.5 T/s

• Logic Style: Standard: Direct answer with minimal scaffolding.

• Code Style: Functional: Standard implementation.

• Summary Style: Conventional: Standard paragraph form.

gemma4:26b-mlx

• Avg. Latency (s): ~5.6s

• Speed: 42.8 T/s

• Logic Style: Analytical: Fast, sharp, and logically sound.

• Code Style: Efficient: Clean, optimized code.

• Summary Style: High-Fidelity: Strikingly accurate/concise.

qwen3.8:27b-mlx

• Avg. Latency (s): ~7.0s

• Speed: 18.7 T/s

• Logic Style: Minimalist: Single-word/Two-word responses.

• Code Style: Basic: Single-function solutions.

• Summary Style: Brief: Zero-frill output.

Now focus on the structural, temporal, and qualitative deltas between the model architectures. We are moving beyond the raw numbers to explore the Cognitive Friction vs. Information Density trade-off observed during the session.

1. The Latency Landscape: Compute-Intensive vs. Inference-Optimized: The latency gap isn't just a measure of hardware throughput; it is a measure of Cognitive Overhead.

The "Thinking" Tax (High-Latency Profile): The qwen3.5:122b model demonstrates a high "Thinking Tax." Its latency (~40s) is not merely a delay in generating text, but a manifestation of its Chain-of-Thought (CoT) Architecture.

• Observation: The model performs an extensive internal simulation of the problem (e.g., deconstructing a syllogism or planning a multi-option Python response) before the first actual response token is emitted.

• Implication: This is "Deep Reasoning." The latency is a signal of computational depth. It is not a "slow" model; it is a "deliberative" one.

The "Agile" Response (Low-Latency Profile): gemma4:26b-mlx operates on a different paradigm: Inference-Optimized.

• Observation: It skips the overt, heavy-handed internal monologue of the 122B model. Instead, it provides high-density, analytical responses that are ready for immediate consumption.

• Implication: This model is optimized for the Human-AI Feedback Loop. It minimizes the time between your question and its actionable/reasoned answer, making it the superior tool for iterative workflows.

2. Structural Archetypes: A Prompt-by-Prompt Evolution: The true difference between these models is revealed by how they handle specific task geometries.

Phase I: Logic & Deductive Reasoning

• qwen3.5:122b: Operas as a Formal Logician. It performs a full deductive deconstruction (e.g., "Premise 1... Premise 2... Conclusion..."). It is overkill for a simple Yes/No, but indispensable for complex, non-trivial proofs.

• gemma4:26b-mlx: Acts as a Sharp Analyst. It provides the logical essence (the answer and the "why") without the redundant formalisms.

• qwen3.8:27b: Acts as a Static Lookup. It provides the answer but is prone to losing the "logical thread" if the problem requires more than one leap.

• qwen3.6:35b: Operates as a Predictable Scaffolder. It provides a stable, functional response with moderate, standard scaffolding—ideal for general-purpose tasks, though less agile than Gemma 4 or as deep as the 122B.

Phase II: Code Generation & Algorithmic Complexity

• qwen3.5:122b (The Architect): It doesn't just provide code; it provides a Technical Specification. It presents multiple strategies (Iterative vs. Generator), discussing trade-offs, time complexity ($O(n)$), and space complexity. This is the model you use for architectural planning.

• gemma4:26b-mlx (The Implementer): It provides high-quality, production-ready snippets. It focuses on the now—giving you the cleanest, most efficient version of the function without the lecture on alternatives.

• qwen3.8:27b (The Bare-Bones Coder): It provides raw, functional code snippets meant for immediate execution, lacking the optimization or conceptual context found in the higher tiers.

• qwen3.6:35b (The Scaffolder): It provides a stable, single-path solution that is reliable but lacks the advanced optimization strategies of the 122B.

Phase III: Creative & Instruction Following

• Creative (Poetry): The 122B model creates Atmosphere (discussing the "red light" and "desolate setting"), whereas Gemma 4 creates Imagery (the "dust on the lens"). The 122B is a writer; Gemma 4 is a poet. Qwen3.6:35b provides Narration (standard, descriptive prose), and Qwen3.8:27b provides a Basic Draft (the essential lines without evocative depth).

• Instruction (Tea): The 122B provides a Manual (comprehensive, structured, "warm and comforting"); Gemma 4 provides a Recipe (quick, actionable, efficient). Qwen3.6:35b provides a Guide (standard, approachable steps); and Qwen3.8:27b provides a Sequence (direct, utilitarian steps).

3. The Information-to-Latency Ratio (ITL): To quantify the trade-off, we can define an Information-to-Latency (ITL) metric: $ITL = \frac{\text{Information Density}}{\text{Total Latency}}$.

gemma4:26b-mlx

• ITL Profile: High ITL

• Qualitative Description: Maximum Value/Second. High information density with minimal waiting.

• Best Use Case: Active Workflows (Coding, Chatting, Interaction).

qwen3.5:122b

• ITL Profile: Low ITL

• Qualitative Description: Maximum Depth/Second. High value per concept, but very low value per second.

• Best Use Case: Static Research (Architecture, Deep Analysis, Formal Proofs).

qwen3.8:27b

• ITL Profile: Minimal ITL

• Qualitative Description: Maximum Speed/Minimum Depth. Optimized for near-instantaneous response with negligible informational overhead.

• Best Use Case: High-Volume Automation & Basic Execution.

qwen3.6:35b

• ITL Profile: Mid ITL

• Qualitative Description: The Standard Baseline. Predictable, moderate value, moderate wait.

• Best Use Case: General Purpose Utility.

Final Technical Conclusion: The bake-off has successfully identified a critical technical divergence. Gemma 4 is your "high-frequency/low-latency" engine. Qwen3.6:35B is your "mid-frequency/stable-state" tool. Qwen 122B is your "low-frequency/high-deliberation" processor. Qwen3.8:27B serves as your "high-velocity/low-complexity" utility. Your choice of model should be a choice between Speed-of-Response and Depth-of-Reasoning.

---

3. Key Observations on Speed

1. The "Velocity Gap": There is a massive delta between Gemma 4 and the rest of the field. Gemma is roughly 2.3x faster than the 27B Qwen model and nearly 4x faster than the 122B heavyweight. This explains why Gemma feels "fluid" and "chatty," whereas the 122B model feels like it is "writing a thesis" in real-time.

2. Scaling vs. Speed: We see a clear inverse correlation between model parameter count and throughput:

• The 122B model is the bottleneck. Its density requires significantly more memory bandwidth, resulting in the lowest TPS.

• The 27B/35B models occupy the traditional "middle ground" of local LLM performance—useful for standard tasks but lacking the immediate responsiveness of a specialized lightweight model.

3. Practical Implications for your Workflow:

• If you are using the terminal or scripting: Use the Gemma 4 or Qwen 27B/35B models. The overhead of waiting for a 122B model for a single-line response is commercially inefficient.

• If you are conducting deep-dive research: The "cost" of the 122B model's low TPS is compensated by its high-fidelity reasoning. You aren't paying in time; you are paying in thought.

---

4. Intelligence Archetypes

1. The Heavyweight: qwen3.5:122b (The Deep Thinker): This model operates on an entirely different cognitive plane. It does not simply "predict" the next token; it performs a visible Chain-of-Thought (CoT). Behavior: Before providing an answer, it "thinks" through the problem. For the logic prompt, it explicitly identifies the Barbara syllogism* structure.
Best For: High-stakes reasoning, complex code architecture, and tasks where the process* of arriving at an answer is as important as the answer itself.

2. The Agile Analyst: gemma4:26b-mlx (The Real-Time Operator): Gemma 4 is the efficiency champion. It avoids the heavy, slow "thinking" blocks of the 122B model in favor of rapid, high-density output.

• Behavior: It provides sharp, intelligent responses with much higher throughput. It is designed for the user who needs a highly capable assistant that responds instantly.

• Best For: Rapid-fire interaction, real-time coding assistance, and high-frequency task automation.

3. The Traditionalists: qwen3.6:35b & qwen3.8:27b-mlx (The Steady State):

• qwen3.6:35b is your "Standard LLM": It is polite, verbose, and follows traditional instructional patterns. It is the "safe" choice for general-purpose tasks.

• qwen3.8:27b-mlx is the "Utility" model: It is stripped of all fluff. It is designed for speed and precision where no nuance is required.

---

5. Anomalies

While most of what you’re seeing is squarely typical of model vs. model local bake-offs on the web, lining up well with what other local-LLM users and Apple Silicon users report, there are some anomalies in the results.

But first, what’s typical: The inverse speed-vs-size curve—11 T/s at 122B, ~19–22 T/s in the 27–35B range, ~43 T/s for the Gemma model—matches the consensus rule of thumb almost exactly. Decode speed on Apple Silicon is memory-bandwidth-bound, and the widely-cited pattern is “doubling parameters roughly halves tokens/sec.” Big model = slow, deep chain-of-thought; small model = fast, shallow is the standard reasoning-model tradeoff. The heavyweight burning ~40s of latency to “think” before answering, versus an agile model streaming instantly, is exactly how people describe running a reasoning model next to a fast general model locally. Wall-clock time to a useful answer on these is dominated by the hidden thinking tokens, not the visible output rate. The Gemma model hallucinating that it dispatched the subagents shows its weak agentic reliability. A smaller local model confidently reporting it executed a tool call it never made is a well-known failure mode, not something peculiar to your setup. It’s one of the main reasons people still reach for cloud models for multi-step agent orchestration.

What’s anomalous:

• MLX does beat GGUF/‎⁠q4_K_M⁠ on Apple Silicon, but the typical, well-measured gap is 15–40% on single-user decode, not the 2x you saw comparing alibaba/qwen3.8:27b-q4_K_M to alibaba/qwen3.8:27b-mlx.

• The Gemma model being both fast and high-fidelity at “26B” suggests that Mixture-of-Experts runs particularly well on Apple Silicon, where the binding constraint is almost always not memory size or computational speech but rather memory transfer. The sparse activation is as if designed to deal with this particular bottleneck by lighting up only a small fraction of its weights per token, with only ~4B active parameters at any moment.

• The ‎⁠qwen3.8:27b-mlx⁠ “single-word/minimalist” behavior is a red flag. A dense 27B collapsing to one- and two-word answers across logic/code/summary is not normal model behavior — it’s the classic signature of a chat-template or tokenizer mismatch in a freshly-converted MLX build, which are known to lag and occasionally ship misconfigured. I’d re-pull the build or check the template before reaching conclusions.

Do note: You’re on a MacBookPro, not a MacStudio. Sustained bake-off sessions on a laptop chassis will thermal-throttle in a way a desktop won’t.

Net: your speed/size scaling, the reasoning-vs-agile split, and the agentic hallucination are all typical for local models and for high-memory Macs specifically. Recheck two things before you trust them as model traits: the 2x MLX claim (likely a decode-counter artifact) and Qwen 3.8’s terseness (likely a template bug). And credit Gemma’s speed to its MoE sparsity, not just its disposition.

---

6. The Trade-off Frontier: Decision Matrix

To optimize your workflow on the MaxxxedOutM5MaxMacBookPro, use the following logic:

1. Does the task require complex logical deconstruction? 👉 Switch to Qwen 122B.

2. Is the task part of a rapid, interactive conversation? 👉 Stay on Gemma 4:26B.

3. Do you need a standard, descriptive explanation without heavy compute? 👉 Use Qwen 3.6:35B.

4. Are you running a simple command-line utility or script? 👉 Use Qwen 3.8:27B.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##going-deeply-into-the-weeds-standing-up-a-new-local-llm-monday-mamlms
##monday-mamlms
##mamlms
##subturingbradbot
##the-llm-bake-off-under-my-dining-room-side-table-the-launch-of-openweight-alibaba-qwen3-8-27b-and-testing-four-local-llms-on-one-maxxedoutm5maxmacbookpro
#going-deeply-into-the-weeds-standing-up-a-new-local-llm
#local-llms
#apple-silicon
#llm-bake-off
#llm-hallucination
#on-device-ai

An autocracy running within a democracy — but one short on the cadre, the discipline, and the goon-captains that made the twentieth century’s tyrannies stick. The country is still a republic; the Republican Party is a personalist autocracy. Vance as Sejanus, Miller drawing up the self-coup, Trump checked out at the center of it:

Bret Devereaux frames the current White House as an autocratic court operating inside a still-functioning democracy.

His most memorable line is that Trump holds two different offices with respect to two different polities:

Trump is President of the United States and King-Archbishop Lord Protector of Republicans…

Share

For he makes the core distinction:

My read, for what it is worth, is the United States, as a country, remains a democracy, but the Republican Party is now a personalist autocracy and brings that nature with it when it is voted into power…

Give a gift subscription

CROSSPOST: BRET DEVEREAUX: At the Autocratic Court of the Chaos-Monkey Trump

<https://bsky.app/profile/bretdevereaux.bsky.social/post/3mti4jadi2k2x>

Bret Devereaux

August 19, 2026

I do not like the fact that this White House very clearly has the structural issues you find in an autocratic court - it is profoundly strange and deeply concerning to basically watch an autocracy running within a democracy. My read, for what it is worth, is the United States, as a country, remains a democracy, but the republican party is now a personalist autocracy and brings that nature with it when it is voted into power.

Trump is President of the United States and King-Archbishop Lord Protector of Republicans. That said, precisely because the imperial court is so deeply, violently dysfunctional, I remain pretty confident that this authoritarian attempt is going to fail and probably fail quite badly. Trump-as-Hoover remains, I think, my modal outcome, though other much, much worse options are possible.

Surely as an historian you appreciate watching how it unfolds from the same damn script every time?

The consistencies are remarkable. I wonder who Sejanus will be (it’s Vance, very obviously Vance)….

To the degree there is a strategy (emotive as it may be) it seems to be focused on keeping control of the GOP and its base, rather than the country.

What worries me is that this is consistent with a gamble that one party, however emaciated, might be enough to hold the country by force.

Also I simply don’t think they have that dog in them, because the main thing that separates modern fascism with the fascism of the previous century is its sheer laziness and a general unwillingness to sacrifice anything for the cause.

And also not having a core cadre who fought in WWI and so already had normalized personal, physical, lethal violence as a behavior pattern. They do not have enough goons and increasingly also seem short on goon-captains.

I think “Trump as Hoover” is increasingly my modal outcome here….

I think there’s a difference here between Stephen Miller, who I absolutely think is probably planning some sort of self-coup and Trump himself who is checked out on most of this and the thing is they can’t actually do the self-coup without the orange man.

That said, it is also consistent with a strategy which is entirely grift focused and recognizes that even if the GOP spends years as a rump minority party, if it is consolidated begin MAGA, the opportunities to grift what remains will be extremely lucrative.

That said, Trump’s approval among republicans has gone from regularly being in the 90s to the 80s and now the 70s and even in some cases the 60s, which may suggest that Trump’s hold over the GOP is not wholly unshakeable if things get wretched enough.

Brad DeLong here: Time to refer back to my Cicero on the last hours of the Roman Republic!

Leave a comment

Read more

The Economist published an unbylined, unsourced piece asserting Daron Acemoglu “counters that [David] Albouy reaches his conclusion by omitting half the data from the original sample.” That claim is simply wrong, and Albouy is right to be angry — no fact-check, no source, no context. Albouy’s actual point is narrow and correct: AJR have no real first stage. Once you correct for clustering, drop 36 conjectured mortality rates, and control for barracks-versus-campaign sources, the mortality–expropriation relationship collapses toward one-in-three significance. The second-stage test statistic isn’t a t-distribution; it’s near-Cauchy — infinite variance, no mean. The IV estimates are unreliable. Moreover, Acemoglu, Johnson, and Robinson ought to be very grateful to David If you take their IV results seriously, the effects implied are embarrassingly and implausibly large: a clock that chimes thirteen. Albouy provides an explanation for what is otherwise a very large implausibility in their story:

One cannot know what to make of this paragraph in the London Economist:

Anonymous: The World’s Most Influential Economist Is Oddly Unconvincing <https://www-economist-com.libproxy.berkeley.edu/finance-and-economics/2026/08/17/the-worlds-most-influential-economist-is-oddly-unconvincing>: ‘David Albouy… showed that some countries were assigned mortality rates borrowed from other[s]…. Correct… and the [Acemoglu] paper’s estimates become unreliable…. Buchner… and colleagues reported that experts they surveyed were somewhat more likely to side with Mr Albouy. Mr Acemoglu… counters that Mr Albouy reaches his conclusion by omitting half the data from the original sample, including on important countries like America, Canada and Australia. It is this combination, along with some statistical choices, that introduces the unreliability, he says. He adds that if he were redoing the paper today, he would make “a number of changes, including in some of the estimation details”—though not to the mortality data…

Share

To start with, the Economist’s lack of bylines makes hit pieces like this one on Daron Acemoglu unconvincing. The lack of sources does as well. Normally, you expect an unsourced “said” or “counters” to be something said to the reporter. But the story quotes a tweet from Noah Smith:

I’ve been yelling about Acemoglu for literally a decade…

Give a gift subscription

And it did not contact Noah. It quotes a podcast segment from Larry Summers:

He leaves out entirely in that analysis the possibility that we will have more rapid scientific progress, more rapid social-scientific progress, or better decision-making because of artificial intelligence…

Share DeLong's Grasping Reality Weblog

without stating the source as well.

Thus I have no idea what the context of the part of the story I take exception to—the statement that Acemogu “counters that Mr Albouy reaches his conclusion by omitting half the data,,, along with some statistical choices…”—comes from. Is this something that Daron said to the story-writer? If so, this is a very bad thing to say. Is the context otherwise? I would like to see the context.

Leave a comment

In any event, that claim is simply wrong. And David Albouy is right to be seriously pissed off:

David Albouy: ‘The funny thing is that the article is about how Daron is isn’t really trusted. And then he ended up making a lie about my work in the article itself. And the reporter didn’t bother fact checking it or running it by me either…

Here is David:

Albouy Colonial Origins

704KB ∙ PDF file

Download

Download

The question is what to make of these two charts:

So let me turn the microphone over to David:

Get 75% off a group subscription

David Albouy: The Colonial Origins of Comparative Development: An Empirical Investigation: Comment <https://pubs.aeaweb.org/doi/pdfplus/10.1257/aer.102.6.3059>: ‘There are several reasons to doubt the reliability and comparability of their European settler mortality rates….

First, only 28 countries have mortality rates that originate from within their own borders. The other 36… are assigned rates based on conjectures the authors make as to which countries have similar disease environments. These assignments are generally unfounded and potentially contradictory.… At a minimum, the sharing of mortality rates across countries requires that statistics be corrected for clustering (Moulton 1990). This correction alone noticeably reduces the significance of the results. If, in the hope of reducing measurement error, the 36 conjectured mortality rates are dropped from the sample, the point estimates relating mortality rates with expropriation risk become substantially smaller, particularly in the presence of covariates, which often gain significance.

Second, the mortality rates never come from actual European settlers…. Instead, the data come primarily from European and American soldiers in the nineteenth century… [some] at peace in barracks… others… on campaign…. Controlling for the source of the mortality rates weakens the empirical relationship between expropriation risk and mortality rates substantially. Furthermore, if these controls are added and the conjectured data are removed, the relationship virtually disappears, suggesting that it is largely an artifact of the data’s construction….

Without a robust relationship between expropriation risk and mortality rates, the AJR IV estimates of the effect of expropriation risk on GDP per capita suffer from weak instrument problems: point estimates are unstable, and corrected confidence intervals are often infinite….

Albouy’s point is this: AJR do not have a first-stage for their instrumental-variables regression. As I put it, when I teach the paper, they report that the association between log mortality, as they assign it, and perceived expropriation risk in the 1980s ( Never mind what extent perceived expropriation risk by a consulting firm evaluating political risk can be understood to be a measure of institutions) passes the null-hypothesis test at 1/1000, at one chance in a thousand. They ought to—because their mortality assignments are carrying information about continents, which carry a lot of information that makes expropriation risk look more salient than it is in the IV—have reported column (4), which passes the null-hypothesis test at 1/25. And by the time one has noted that some soldiers are in barracks and some of the data come from poor laborers, the first-state is down to 1/3. Thus the distribution of their test statistics from the second-stage of their IV regression is not a t-distribution, but is because of the weak-instrument problem near to a Cauchy distribution: that thing that not only has infinite variance and standard deviation, but does not even have a mean.

That is a fair point. The IV estimates are highly unreliable.

After pointing that out, I go back to the OLS correlation between expropriation risk and prosperity today, and we talk about the various ways the world might work that would produce that OLS scatterplot:

<https://pubs.aeaweb.org/doi/pdfplus/10.1257/aer.91.5.1369>

What other than a relationship between trust in the rule of law and the solidity of property rights on the one hand and economic activity leading to prosperity on the other hand could produce this scatterplot?

Refer a friend

Now Acemoglu, Johnson, and Robinson defend the hill of their IV because, according to the rules of economics since the empirical-causal turn, an OLS scatter cannot be interesting or the basis for an AER paper. But their OLS scatter is interesting, and important, and worth talking about.

Daron Acemoglu, Simon Johnson, and James Robinson: Hither Thou Shalt Come, But No Further: Reply to “The Colonial Origins of Comparative Development: An Empirical Investigation: Comment” <https://economics.mit.edu/sites/default/files/inline-files/Hither%20Thou%20Shalt%20Come>: ‘Overall, Albouy’s “Comment” amounts to a series of objections to our approach. All of these objections, upon closer inspection, are far from compelling, are often unfounded, and prove minor and largely inconsequential for the robustness of our results. The big picture from AJR (2001) remains intact and remarkably robust: Europeans were more likely to move to places that were relatively healthy, and when they moved in larger numbers, they imposed better institutions, which have tended to persist from the colonial period to today.

In my view, this rhetorical position by Acemoglu, Johnson and Robinson is a huge mistake. They should be very glad to drop their IV results from the discussion.

Acemoglu, Johnson and Robinson went down the road of their IV because of the potential criticism that causation is not flowing from perceived security of property to prosperity, but rather that prosperity has lots of effects that lead to perceived security of property. They thought that they could use settler mortality to identify a component of perceived expropriation risk today that was plausibly independent of these reverse causation factors. In my view, the right way to understand what they wound up doing is to say that they regressed current prosperity on what perceived expropriation risk today would be if you knew only about settler mortality in the past and nothing else. The slope of that regression—the regression of prosperity today on what one would expect perceived expropriation risk to be from settler mortality—is the hill they are willing to die on.

Their OLS scatterplot gives them a coefficient β-hat =0.522: a one-unit increase in their security-of-property today index leads to an 0.522-unit increase in prosperity today, a 68.5% increase.

And their IV scatterplot gives them a coefficient β’-hat = 0.944, which means a one-unit increase in what you would think perceived expropriation risk is today based on settler mortality leads to an 157% increase in prosperity today.

In the real world, a country like New Zealand with a log prosperity score of 10 and a perceived security-of-property score of 10 was back before 2000 about 5 times richer than an Egypt with a log prosperity score of 7 and a perceived security-of-property score of 7. AJR’s IV results say that the effect of security-of-property on prosperity ought to be much much bigger than that: it ought to be about 20 times richer. And it would be, if the relationship between security of property leading to investments in physical and human capital and in enterprise and innovation were the only things operating, were there no reverse causation by which prosperity leads to social unrest that threatens the security of property. But there is such social unrest, AJR’s IV results tell us. There is powerful reverse causation: a doubling of prosperity sets in motion societal forces that would, if they were the only things operating, reduce your perceived security-of-property score by 0.4.

Now: any claim that prosperity structurally reduces governance quality contradicts an overwhelming body of evidence across multiple disciplines:

1. The Modernization Hypothesis (Political Science): Lipset (1959) argued the that it was economic development that created the social conditions for democracy and good governance—an educated middle class, urbanization, organizational capacity, and norms of civic participation. Przeworski and Limongi (1997) and Boix and Stokes (2003) found strong empirical support that higher income increases the probability of democratic consolidation. Acemoglu et al. themselves, in later work (Journal of Political Economy, 2008), examine this relationship. The literature is not unanimous, but no serious strand of it argues that prosperity destroys governance quality.

2. Historical Evidence on Expropriation and Revolution: Revolutions, expropriations, and institutional breakdown have historically occurred overwhelmingly in poor countries, not rich ones. The Russian Revolution and the wave of postcolonial nationalizations all arose in contexts of poverty, grievance, and institutional fragility, not in contexts of prosperity. The handful of rich-country institutional collapses (Weimar Germany) are exceptions, and even there the mechanism was economic collapse, not economic success.

3. The Resource Curse Literature: The closest empirical phenomenon to is the resource curse: oil wealth sometimes corrodes governance by enabling authoritarian consolidation without taxation. But this mechanism is specific to extractive resource wealth, not to prosperity in general, and it operates through a very particular political economy channel (Ross, 2001; Robinson, Torvik & Verdier, 2006). It cannot be generalized to a structural negative relationship between income and governance across all former colonies.

4. The Cross-Sectional Evidence: Simply looking at the data: the richest former colonies—Singapore, South Korea, Botswana, Mauritius—tend to have better governance, not worse. The poorest—the DRC, Haiti, South Sudan—tend to have the worst. A relationship is not what the scatter plot shows.

Thus the implied backwards causation from prosperity to poor governance quality is not merely theoretically awkward—it is empirically falsified by the entire body of comparative political economy.

This means the only semi-live explanation of the OLS-vs-IV gap is—if we take it as anything other than the Cauchy distribution arriving at the picnic and getting out its refreshments, as an example of play, stupid games and win stupid prizes—that it is the result of measurement error in governance quality attenuating the OLS estimate and pushing it downward. That is the interpretation AJR themselves favor. They conclude that the instrumental-variables strategy does not primarily correct for reverse causation. Rather, it primarily corrects for attenuation bias from mismeasured institutions.

But taking that governance quality is mismeasured gets Acemoglu, Johnson and Robinson into even worse trouble.

Their argument is that we cannot today see the true institutional quality of governance G. Instead we see a corrupted and noisy measure of institutional quality H. But if we look at just that part of H that is correlated with settler mortality, we recover a better measure of the institutions that matter. In order for this explanation to work, however, settler mortality in the age of imperialism has to (a) be correlated with that part of modern-day institutions that matter for prosperity, the G, while also (b) being uncorrelated with those parts of modern-day institutions that do not matter for prosperity. Settler mortality C must be: 1. relevant, in that it is correlated with the component of observed institutions that genuinely cause prosperity; and 2. selectively orthogonal, uncorrelated with the part of observed institutions counted as good governance that do not in fact matter for prosperity.

And my response to this is simply this: You are putting me on.

There is no way that 1800s soldier mortality gives us a better measure of institutional quality today, than does looking around at everything we can see about institutions and constructing high-information measures.

The right response from Asimov, Johnson, and Robinson to Albouy is not to dig in deeper, is not to die on the hill of a weak instrument IV, but rather to be very grateful that Albouy has pointed out the magnitude of the weak instrument problem and thus provided a Cauchy distribution explanation of the weirdly implausibly and embarrassingly large size of their IV coefficient. They really do not want to be on either of the hills that defending their IV could put them on:

1. either prosperity has catastrophically destructive effects on governance,

2. or (ii) colonial-era mortality is a better measure of what matters for prosperity today in modern institutions than modern institutional analyses can produce.

Neither of those is at all defensible.

But Wait! There Is MOAR! Much, Much Moar!!

ADDITIONAL CRITIQUE 1: It’s Human Capital, Not Institutions: Journal of Economic Growth, Glaeser, La Porta, Lopez-de-Silanes, Shleifer 2004: Europeans who settled in healthy places brought themselves—educated, literate people with specific cultural and legal traditions. The AJR IV identifies where Europeans settled, but settler presence = human capital accumulation, not just institution-building. The instrument may be picking up human capital rather than institutional quality.

But: AJR control for fraction of European population directly, and the institutional effects survive. Moreover, human capital and institutions are not cleanly separable — colonists built institutions deliberately. The critique identifies a real confound but doesn’t break the core result.

ADDITIONAL CRITIQUE 2: Geography, Not Institutions: Various, Sachs & al.: Malaria, latitude, and disease burden directly affect economic productivity, not just through the institutional channel. Settler mortality may proxy for the disease environment that directly depresses output today, violating the exclusion restriction.

But: AJR control for latitude, temperature, humidity, malaria prevalence, and more—and the institutional effect survives all of these. The paper’s robustness tables are unusually thorough on exactly this point.

ADDITIONAL CRITIQUE 3: Institutions Don’t Persist That Way: Various, Path Dependence Skeptic Historians & Political Scientists: The persistence mechanism is assumed rather than demonstrated. Colonial institutions were often radically transformed at independence, and many countries have undergone multiple regime changes. Why would 17th-century institutional choices echo so cleanly into 1990s PRS scores?

But: The correlation between early institutions and current institutions is actually quite well-documented empirically. The persistence is real, even if the mechanisms are complex. AJR‘s own later work, i.e., Why Nations Fail, develops the persistence story more carefully.

ADDITIONAL CRITIQUE 4: The Exclusion Restriction Is Untestable: Standard IV Critique: Settler mortality back then affects current income through channels other than “institutions”—by shaping culture, work norms, trust levels, or through direct epidemiological legacies.

But: Untestability is a feature of all IV designs, not a special problem for AJR. And AJR‘s robustness to controlling for an unusually long list of potential direct channels is about as good as IV evidence ever gets.

Bottom line (for teaching): The Albouy data critique is the one that genuinely stings—not because it refutes the finding, but because it raises legitimate questions about fragility. On the other hand, the very large size of the estimated IV coefficient is very embarrassing and suggests that something has gone catastrophically wrong with the analysis. Albouy provides a way out of that very legitimate critique.

The others are important conceptual challenges that AJR largely anticipated and addressed. The paper’s core claim—that colonial history explains a large fraction of income differences today with “institutions” as a primary channel—has proven surprisingly durable; while also remaining puzzling and, in a sense, unbelievable.

Memo to self: <https://datahub.berkeley.edu/user/jbdelong/lab/workspaces/auto-1/tree/working_20251227/2026-04-08-DELIVERED-econ-196-week-9-reversals-of-fortune.ipynb>

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##‎⁠wait-what-mr-albouy-reaches-his-conclusion-by-omitting-half-the-data-from-the-original-sample-another-chart-of-the-day
##enlarging-the-bounds-of-human-empire
##public-reason
##chart-of-the-day
##‎⁠that-is-simply-wrong
#wait-what-mr-albouy-reaches-his-conclusion-by-omitting-half-the-data-from-the-original-sample
#colonial-origins
#acemoglu-johnson-robinson
#david-albouy
#weak-instruments
#institutions-and-growth
#reversals-of-fortune
#economic-history
#cross-country-economic-growth

Every assignment produces two things: the paper turned in now, and the judgment the student builds for later. AI can raise the first while hollowing out the second—and only one shows up in the grade. So I need to redesign my courses around that fact, and that is only one part—the assessment part—of how “AI” has already come for American higher education:

I am not teaching at all in this forthcoming semester. For reasons that nobody has explained to me and that I see no point in trying to dig into, if I teach anything at all this semester, then: my health insurance turns from the gold-plated grandfathered-in employer-sponsored health insurance of someone hired by the University of California in the 1990s into the much skimpier employer-sponsored health insurance the university currently offers to new temporary employees.

Share

But I am going back into the teaching rotation in the spring of 2027 for Econ 210a and Econ 135: a 25-person and a 75-person class. and I will then have to face a system where information-technological disruption is currently producing outcomes like this:

Give a gift subscription

Via Paul Novosad <https://x.com/DKThomp/status/2090076039141552443>, who writes:

Every college syllabus should include these graphs. Use AI for homework, you will get it done faster and get a higher grade, and then get crushed on the exam…

Share DeLong's Grasping Reality Weblog

And Derek Thompson connects the dots with respect to assessment:

Derek Thompson: ‘Unless testing shifts toward in-person and/or blue books soon, AI is going to turn a whole lot of education into the informational equivalent of "wow I [paid a guy, who] squatted 200 lbs at the gym yesterday, new personal record!"…

Leave a comment

He is correct.

At least in my estimation, every piece of written work that we assign and expect to be turned in needs to have a 10-minute one-on-one with either me or one of the TAs talking and arguing about the document produced. This will do two things:

• First, it is something that we ought to have been doing all along, but because we are lazy, we developed an educational model with insufficient feedback, engagement, and dialogue.

• Second, it is the only way to try to avoid the graph pattern above that students, as they ask the AI to do their work, are thinking: I will have to explain this to the professor or the TA next week.

Get 75% off a group subscription

75 students. 1 TA. That is two of us. 10 minutes of assessment/socratic dialogue per student per assignment. That is 12 and a half hours of our time for one assignment. Could we get away motivationally with only checking in a third of the time, and choosing the check-in at random? Maybe that would then be three hours per assignment. What with slack and so forth, that would be one full workweek of the 10 workweeks of time devoted to the course by the teachers. That is, I think, doable.

That is a plan: 10 assignments, 3 check-ins in person per student.

I think I need to stress here that, while this is a substantial increase in workload that will crowd-out other things, this shift in the mode of assessment is not so much a problem as a problemtunity.

First, the something it crowds out is, on inspection, largely the low-value grading of artifacts whose provenance was sometimes uncertain, and in which the feedback loop that would both enable and force student improvement was not closed. We told ourselves for decades that we were providing sufficient feedback in sections and in paper comments, and mostly we were not.

There is a load-bearing confession here: We built the modern large-lecture university course as a machine for economizing on faculty attention. The problem set turned in, the blue book graded, the term paper marked up in the margins and handed back to a student who looked only at the letter at the top. This was never pedagogically optimal. This was faculty effort-minimizing. We dressed up a budget constraint we had constructed for our convenience as a philosophy of education.

So the one-on-one is not a defensive crouch against cheating. Framing it that way concedes the wrong ground—it makes the professor a border guard and the student a smuggler, which is miserable. When I sit across from a student and we argue for ten minutes about the causal story in her paper—why she chose this mechanism and not that one, what evidence would have changed her mind, where the awkward document in the archive doesn’t fit—I am measuring the stock of judgment directly, at the source, rather than inferring it from a proxy that has just been debased. The AI can produce the draft. It cannot sit in the chair and be caught not knowing why the draft says what it says.

Here is the part that makes it a problemtunity rather than a problem. The behavioral effect runs backward through time. A student who knows that next week she will have to explain this to the professor or the TA is, this week, a different writer. Use the machine to lower the cost of the prose and to supply criticism, not to supply the judgment; keep verification in your own hands, because the person in the chair across from you is going to check the chain from claim to evidence. Not “you must write every sentence yourself,” which is a losing and slightly ridiculous war to fight. Rather: “you must be able to think, in real time, about the sentences that appear under your name.”

That is a better commitment device than the blank page ever was, because it targets the muscle and not the ritual.

And it is, frankly, the education I myself did want to have received, and did kinda-sorta. The tutorial—Oxbridge’s expensive glory, which the American research university abandoned as a luxury it could not scale—turns out to be the assessment mode that a world of cheap text forces back upon us.

When AI writes the prose; that frees the hour. Spend the freed hour in the chair, arguing. That is not a tax on teaching. That is teaching, finally unmasked as the thing it was always supposed to be.

One workweek out of fifteen. I’ll take that trade.

Share

And now I have to figure out how to integrate the coming of AI for summarization and search with respect to the reading, with respect to writing prose, and with respect to the programming tasks that I hope to assign come next spring.

Refer a friend

I see this AM that Johan Fourie has some thoughts about prose-writing that are relevant:

Johan Fourie: Writing Is Not Thinking <https://johanfourie.com/files/wp/JF_WritingIsNot_v3.pdf>: ‘Writing technologies lower the cost of producing candidate text much more than they lower the cost of judging which candidate to keep. Text has become cheap. Judgment has not. A model that returns criticism may help with judgment, as Section 4 explains, but the asymmetry drives the results….

1. Search. Cheaper drafting improves the argument ultimately selected….

2. Human capital. When generated text replaces the practice that forms judgment, assisted output rises today while unaided capability falls tomorrow. Gaps in performance narrow as gaps in unaided capability widen….

3. Reference points and defaults. An author who is uncertain… and first reads a generated draft… will find the draft is a plausible default… and stop [thinking] earlier[—too early, in fact]….

4. Signalling. When fluent prose becomes cheap to produce, English fluency becomes a less reliable signal of quality. Wherever poor English had concealed good evidence… [improving] it makes prose more informative….

5. External effects. If many authors draw on shared generative defaults, individual gains occur alongside field narrowing: more material to screen, but fewer distinct questions and explanations under study…

Share

If I understand where he is coming from and where he gets to sufficiently:

• The largest, cleanest gains from AI go to second-language researchers, because the cost of English is unrelated to the quality of their evidence — removing it is pure benefit.

• The dangerous move is not using AI but using it first: forming the representation is the act that both makes the manuscript original and builds the author’s judgment.

• The defensible workflow is: own account → model for language and criticism → independent verification; disclosure should name the task delegated, but disclosure is a cheap message, so journals should police the claim-to-evidence chain instead.

With this being his biggest worry, I think:

The historian from the opening still has her awkward document. If she writes her own account first, the model can improve it and lower the cost of its prose. If she reads the model’s account first, the document may never disturb the familiar story. No finished manuscript reveals the difference between those two mornings of work…

And the principal objection to his argument as a whole being the one set out by Deirdre McCloskey back in 1985. As Fourie puts it:

Fluent synthesis may… conceal a broken link between a claim and the world…. Workflows… [have] different access to evidence and different incentives to verify it. [Plus] the framework of this article… faces a deeper objection… stated by… McCloskey (1985) [who] rejected the premise that content and expression are separable. In production terms, scholarship cannot be written as the sum of two functions. An author discovers the details of argument only by writing them out, and in doing so often discovers a flaw in its foundations…

Plus I thought this was nicely put, although calling ten kilos—six miles—a “short distance” run leaves me impressed. When I think “short distance”, I think half a mile:

Johan Fourie: What I Think About When I Think About Writing <https://www.ourlongwalk.com/p/what-i-think-about-when-i-think-about>: ‘I run short distances: usually ten kilometres, in Jonkershoek, in the early morning as the sun rises over the peaks. It’s spectacularly beautiful. These are also the best times to think. I think about… research. My family. Blog post topics. More research ideas…. Why has no one founded a company like that? A bug walking across the road…. I need to remember to renew the family Wildcard, otherwise I’ll have to pay to get into Jonkershoek. Oh, and the car licence….

You get the point. Which is why I find it fascinating when people say that ‘writing is thinking’. No, writing is sitting down and forcing yourself to think about something…. It is not the writing that does the thinking. It is the thinking that does the thinking. You don’t need the writing to do the thinking, but the discipline of the writing helps the thinking, yes…. AI need not displace thinking if we are careful about the sequence….

‘Writing’ is not one activity…. Turning a decided thought into a grammatical sentence is one task. Taking a vague, half-formed, slightly embarrassing impression and forcing it into words explicit enough to inspect is a completely different one. So is choosing which causal story to tell…. [So is] checking whether the archive supports it. Delegating the first is… sometimes a mercy. Delegating the second is where the trouble starts…. If drafting… takes seven hours and checking it against the evidence takes one… but a model drops the drafting to one hour… the tool has quadrupled the number of arguments I can test…. As drafting approaches free, the number of versions I can examine… approaches my own capacity to judge them. That capacity has not improved at all.

That capacity is the paper’s second output. Every piece of writing produces two things: the manuscript you publish now and the judgement you will develop. Judgement is human capital, a stock built by costly practice and lost through disuse….

The subtler loss is the blank page…. Before [AI text] generation… nobody had to design a commitment device for writing. The blank page was one. No effort meant no manuscript…. A generated draft changes the zero-effort outcome from a blank page to a plausible paper….

What matters is not whether you can write every sentence yourself. It is whether you notice something, see why it might matter, follow the idea, and know when the answer is wrong…

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##here-in-academia-ai-has-arrived-for-software-coding-for-summarization-search-for-writing-prose-for-assessment-chart-of-the-day
##academia
##public-reason
#mamlms
#subturingbradbot
##chart-of-the-day
##text-got-cheap-judgment-didnt-how-i-hope-to-teach-in-2027-ai-has-arrived-for-coding-search-prose-assessment-three-opportunities-and-one-problemtunity
#here-in-academia-ai-has-arrived-for-software-coding-for-summarization-search-for-writing-prose-for-assessment
#ai-in-academia
#writing-is-not-thinking
#mamlm-academic-problemtunity
#socratic-check-in-in-person-assessment
#stopping-thinking-too-soon
#teaching-in-2027
#johan-fourie

A “top-five” result apparently undone by a single inverted variable and thus a division where there should have been a multiplication: what one coding slip says about economics. Fix the inversion and a thousand-times-cited AER finding loses its significance, its sign, and its story, as Joe Francis spends his time performing the public-good service of making the machinery of knowledge actually work:

Joe Francis is doing more replications.

He finds that Galor and Özak’s claims on “the agricultural origins of time preference” <https://www.aeaweb.org/articles?id=10.1257/aer.20150020> attain statistical significance only because their “area weights” variable is wired into the analysis exactly backward: instead of downweighting small areas that people would find it easy to move away from, they downweight large areas where people are more likely to stick.

You have a prior—a strong, theoretically motivated, Malthusian-selection prior that patience got culturally bred into populations where the agronomic return to waiting was high. You run the specification. It comes out significant and correctly signed. And here is the trap that catches all of us who are even passable Bayesians: you do not audit results that confirm what you already believe. You audit the surprises. Confirmation is where your guard is down. Oded and Ömer divided where they should have multiplied, the number came out the color they expected, and nobody re-derived the weight variable because nobody had a reason to. It happens. It has very nearly happened to me.

Thus: “The estimator gives the largest weight to the smallest regions. A tiny 34 km² Swiss region receives 170,000 times the weight of the largest region. The influence of small, urban regions where internal migration is most severe is thereby maximized, making the problem that area weights were supposed to address worse…”

And once you wire things in the right way, the coefficient loses statistical significance completely and loses its sign: it is not the case that higher potential crop yields and shorter growth cycles that rewarded delayed gratificatication are associated with the development of a more patient human cultural matrix, but rather the reverse.

The broken academic incentive structure is Francis’s deeper target: even a demonstrable error won’t trigger retraction or correction, because the “top five” journals essentially never do so.

The Poor Rich World

This is a blog on economic history.

By Joseph Francis

CROSSPOST: JOE FRANCIS: Running Out of Patience with Economics

The Poor Rich World

Running Out of Patience with Economics

When doing post-publication peer review, it is easy to become pessimistic about the whole human enterprise of knowledge production. It is easy to run out of patience due to the errors one finds. Furthermore, even if I find a fault in one of the articles published in the “top five” economics journals, I know that it will not be retracted or even corrected…

Read more

3 days ago · 9 likes · Joseph Francis

<https://poorrichworld.blog/p/running-out-of-patience-with-economics> <https://poorrichworld.blog>

A Strange Case of Inverted Area Weights

Joseph Francis

Aug 18, 2026

When doing post-publication peer review, it is easy to become pessimistic about the whole human enterprise of knowledge production. It is easy to run out of patience due to the errors one finds. Furthermore, even if I find a fault in one of the articles published in the “top five” economics journals, I know that it will not be retracted or even corrected.

The article “The Agricultural Origins of Time Preference” (2016) by Oded Galor and Ömer Özak is a case in point. Published in the American Economic Review, the article presents a grand narrative of persistence. Galor and Özak hypothesize that “geographical variations in the natural return to agricultural investment generated a persistent effect on the distribution of time preference across societies” (p. 3065). Higher potential crop yields and shorter growth cycles, they say, rewarded delayed gratification, so a culture of long-term orientation was selected for over generations and persists in populations today.

The article has garnered over 1,000 citations on Google Scholar, but it is in no way, shape, or form robust.

The evidence for this narrative comes from a simple correlation between individuals in the World Values Survey and historical FAO data on potential crop yields and growth cycles in the places of their ancestral origins. Macro-historical regressions across countries are, however, famously vulnerable to confounding factors such as geography, modern national institutions, or contemporary culture, which can easily drive the correlation.

Galor and Özak therefore run a version of their regressions with country fixed effects. Individuals are compared within each country, and national institutions drop out. In the published article that specification is statistically significant: historical crop yields and patience, across 1,356 regions.

But the problem is that people have moved. The agricultural characteristics of a person’s current location might not then reflect the environment that supposedly shaped their ancestors’ time preference.

They claim to deal with this by using area weights. As they put it, “observations are weighted by the scale of each region, mitigating the effect of internal migration” (p. 3097). Larger regions, on this logic, are less susceptible to internal migration crossing their borders, so assigning them higher weights should reduce this measurement error.

Nonetheless, their code then does the opposite.

For weights, it uses a variable called “invwvsarea.” When that variable is reverse engineered, it turns out that it is in fact area weights inverted. Why they did this, is not clear, but it has the effect of aggravating the issue that they claim to be addressing. By weighting by the inverse of the area, the estimator gives the largest weight to the smallest regions. A tiny 34 km² Swiss region receives 170,000 times the weight of the largest region. The influence of small, urban regions where internal migration is most severe is thereby maximized, making the problem that area weights were supposed to address worse.

When the actual, non-inverted area weights are used instead, the coefficient for crop yield drops from 0.032 (p=0.010) to an insignificant −0.033 (p=0.39), as shown in Table 1. Indeed, all their main results essentially disappear. And the same happens when no weights are used at all: the coefficient falls to 0.001 (p=0.85), indistinguishable from zero. Their results are artifacts of doing the opposite of what is described in the text.

Yet this probably does not matter. Even though the article is not robust, it will not be retracted, or even corrected. That will not be possible because the “top five” economics journals do not do such things. As the Economist recently observed, we know that things are much better in economics than in the rest of the humanities and the social sciences because the “five leading journals have seen just four withdrawals in their combined 570-year history.” Their peer review is so robust that it would be impossible for an article like this to be published in the first place. Consequently, those area weights cannot really have been inverted.

Further Reading

The article examined in this post is:

• Galor, Oded, and Ömer Özak. “The Agricultural Origins of Time Preference.” American Economic Review 106, no. 10 (2016): 3064–3103. Data and replication archive.

Replication Files

The replication package for this post is available here.

<https://poorrichworld.blog/p/running-out-of-patience-with-economics> <https://poorrichworld.blog>

The Poor Rich World

Running Out of Patience with Economics

When doing post-publication peer review, it is easy to become pessimistic about the whole human enterprise of knowledge production. It is easy to run out of patience due to the errors one finds. Furthermore, even if I find a fault in one of the articles published in the “top five” economics journals, I know that it will not be retracted or even corrected…

Read more

3 days ago · 9 likes · Joseph Francis

The Poor Rich World

This is a blog on economic history.

By Joseph Francis

Brad DeLong here: I part company with Joe is the counsel of despair—the bleak “it will never be retracted, so none of this matters.” I don’t buy it. Retraction is the wrong metric. The right metric is what the profession believes and builds on, and there the machinery works, albeit maddeningly slowly and expensively.

It is true, to highlight what Joe writes:

When doing post-publication peer review, it is easy to become pessimistic about the whole human enterprise of knowledge production. It is easy to run out of patience due to the errors one finds. Furthermore, even if I find a fault in one of the articles published in the “top five” economics journals, I know that it will not be retracted or even corrected…. [Oded Galor and Ömer Özak’s] results are artifacts of doing the opposite of what is described in the text. Yet this probably does not matter. Even though the article is not robust, it will not be retracted, or even corrected. That will not be possible because the “top five” economics journals do not do such things…

Nevertheless, I think Joe is wrong to despair here. Yes, it looks like Oded and Ömer divided where they should have multiplied—it happens—and did not catch it because the results were as they expected and, as good Bayesians, you don’t spend that much time checking expected results—that also happens. But things that do not replicate are weeded out, albeit much more slowly than they should be. And people do have a very healthy skepticism about results that are close to the edge of statistical or economic significance, or that cannot be shown to rely on a strong solid correlation.

It is not just computational error. It is also that, say, you have ten yes-no decisions to make in running any empirical analysis that could go either way. What you should do is to choose the way that makes most sense to you, and run along several tracks where things are debateable. But there is the temptation to choose the track that gets you closer to the result you think is sensible. Succumb to that, and you wind up with the strongest of 1024 possible results. Alternatively, get yourself checked by somebody hostile for ideological reasons, and they can come up with the weakest of 1024 possible results as a way of dismissing your analysis.

Stepping back, the persistence literature keeps reaching for a lever—settler mortality, ancestral crop yields, ruggedness, caloric suitability—that will let a cross-sectional correlation masquerade as a causal parameter, without a structural model that says how the past reaches into the present.

It turned me into a Heckmanite.

I do not believe you have earned the right to say “causal” unless you can write down the structural model and the story about why the identifying variation is even remotely exogenous. “Deep roots” regressions almost never do. The area weight was a half-hearted patch over the wound that people can and do move to opportunity. Inverting it tore the patch off and rubbed salt in.

But things that do not replicate do get downgraded; they stop anchoring dissertations; the second generation of citations turns sour. It is a shame it takes a decade and a Joe Francis rather than an afternoon and a journal’s own referees. We should be able to do much better.

Which is why we need preregistration and replication packages. And that is why Joe Francis undertaking this mission is a very good thing for the profession as a whole.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##crosspost-joe-francis-running-out-of-patience-with-economics
##public-reason
##crosspost
##joes-subheadline-a-strange-case-of-inverted-area-weights-but-i-mostly-want-to-praise-joe-francis-for-spending-so-much-of-his-time-on-post-publication-peer-review
#joe-francis-running-out-of-patience-with-economics
#joe-francis
#running-out-of-patience-with-economics
#post-publication-review
#agricultural-origins
#replication-crisis
#garden-of-forking-paths
#persistence-studies
#heckmanite
#causal-inference
#preregistration
#replication-packages
#economic-history
#confirmation-bias

The blow-up of Leopold Aschenbrenner’s Situational Awareness is a sign that we are near the peak of the bubble: this is what it looks like from the inside when demand stops riding on industry fundamentals and starts riding on leveraged-buyer reads of market sentiment. It is a sign that positive-feedback investment strategies are rampant and that demand curves are starting to slope the wrong way:

During an asset market boom, even a euphoric boom, prices rise as good news arrives and as more and more people take their money and decide that this is indeed the wave of the future. But then increases in demand switch from people becoming aware of the opportunity and bringing their money in, to people willing to bet on rising prices, willing to ignore Risk Management 101, and eagerly leveraging-up and pouring that leverage into the booming asset class. Rising prices taken as a reason to buy more induce a situation in which the smart money starts leaving quietly, and the game silently switches from owning value to finding a greater fool before everyone else does.

Share

And then people are on the edge between hanging on hoping a greater fool comes along and selling out now. The shape of demand no longer rides predominantly on expected developments of industry, but on various reads of market sentiment.

That period in which a lot of people are following positive-feedback investment strategies is a period of “distress”.

We know it is distress because then we get things like this from the most overleveraged:

Give a gift subscription

Via Adam Tooze <https://adamtooze.substack.com/p/top-links-1199-big-losses-meeting>, FT Alphaville <https://www.ft.com/content/340bf9e7-0e67-4d19-b671-3dc8186efb99> picking up “a cool table on Wikipedia” <https://en.wikipedia.org/wiki/List_of_trading_losses> based on Tom Coleman <https://rpc.cfainstitute.org/sites/default/files/-/media/documents/book/rf-publication/2011/rf-v2011-n3-1-pdf.pdf>.

The write-up is by Toby Nangle:

Toby Nangle: A Leaderboard of the Biggest Trading Losses of All Time <https://www.ft.com/content/340bf9e7-0e67-4d19-b671-3dc8186efb99>: ‘We found a cool table on Wikipedia.. to more easily contextualise the quantum of Aschenbrenner’s loss…. There are two steps to a fund losing a lot of money. The first step is to inspire faith in either a large number of fairly wealthy people or a small number of immensely wealthy people…. The second step is to do [is]… throw together a credible investment thesis, [and] have sufficiently high conviction… to cast aside risk management 101, maybe chucking a bunch of financial leverage into the mix…. The second step is easy. There are thousands of people yoloing in their mums’ basements around the world doing just this right now…. So if you’re interested in maximising your ranking on any quantitatively measured global leaderboard of trading losses, the first step is probably more important….

We’re fairly sure that this league table, like every other we’ve chanced upon, is only really capturing the kind of meltdowns that make good copy. There’s Tiger Global’s ca. $40bn bloodbath in 2022, for example, which arguably should put it at the top of the list. There’s a case that Jane Street…[belongs] given the reported $15bn hit that it took in July from its exposure to Aschenbrenner…

Leave a comment

That last is a reference to:

Jill R Shah & Joshua Franklin: Jane Street Suffers $15Bn Hit After Meltdown at Situational Awareness: Jane Street posted a roughly $15bn loss in July after turmoil at AI-focused hedge fund Situational Awareness wrongfooted the US trading firm. The New York-based firm disclosed the figure to lenders as part of a deal to shift its roughly $11bn public debt pile to private investors including Pimco…. Jane Street has generated more than $40bn in net trading revenues in the year to Friday, even accounting for the July loss, which exceeds its entire haul for 2025…. Jane Street’s investment in Situational Awareness was unusual because the firm trades its own capital.… a former Jane Street employee worked at Situational Awareness and Jane Street co-founder and partner Robert Granieri attended [Leopold] Aschenbrenner’s wedding in California in recent weeks….

Jane Street was established in 2000 by a small group of founders including Granieri, who previously worked at Pennsylvania-based Susquehanna. It uses technology to make markets across assets such as equities, bonds, exchange traded funds and more. In recent years, it has expanded into longer-term strategies as well as investments in private companies, including AI lab Anthropic and data centre operator CoreWeave…

Share DeLong's Grasping Reality Weblog

Time to pull out the Kindleberger! This time, from A Financial History of Western Europe:

At some stage in the process it becomes clear to a few, and then to more, that the fallacy of composition is at work, that the whole is rather less than the sum of the parts, that credit positions are extended beyond some limit sustainable in the long run, and that maintenance of capital gains depends on getting out of assets rising in price ahead of others.

There follows a period of what may be called ‘distress’: ‘We have no crash at present, only a slight premonitory movement of theground under our feet,’ wrote Lord Overstone to his friend, G. W. Norman, on 1 November 1845 (O’Brien, ed., 1845 [1971], Vol. 1, p. 368). From time to time the distress abates. On other occasions it intensifies. More and more speculators seek to get out of whatever was the object of speculation, to reduce their distended liabilities, and switch into money; and more and more it becomes clear that not everyone can do so at once.

There is a rush, a panic, and a crash—or perhaps a lender of last resort intervenes to make clear that it will furnish the market all the cash it insists it requires. In this circumstance, perhaps belatedly, panic and distress subside…

Get 75% off a group subscription

The phases of the process are: displacement—a technological or a super-political shock that calls forth a need for real economic adjustment and change—diffusion of euphoria as adjustment takes place, distress, crisis, panic, and then—perhaps—a lender of last resort.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##‎⁠bubble-watching-this-is-what-happens-in-the-type-of-period-called-distress-chart-of-the-day
##macro-outlook
##finance⁠
##‎⁠chart-of-the-day
##‎⁠according-to-kindleberger-things-like-the-situational-awareness-colllapse-of-leopold-aschenbrenner-happen-near-the-peak-of-a-bubble-they-are-a-sign-of-the-shape-of-demand⁠
#bubble-watching-this-is-what-happens-in-the-type-of-period-called-distress
#bubble-watching
#kindleberger
#distress
#financial-crisis
#manias-panics-crashes
#leopold-aschenbrenner
#situational-awareness
#jane-street
#positive-feedback-investment-strategies

OpenAI added a billion dollars of revenue and three billion dollars of losses in a single quarter, and now has apparently paused at least model training for reasons. When a firm best at generating vibes starts conserving cash instead, the bubble peak is likely to be close. I am surprised: Claude Code and OpenAI Codex are close enough to be a matter of taste, yet Anthropic’s revenue nearly tripled last quarter while OpenAI’s grew only (only!) 18%:

The shift from equity to debt financing of datacenters has been one powerful sign of an approaching bubble peak. Gary Marcus reads OpenAI’s pausing of some aspects as model training as a second sign: husbanding cash now appear to be more important for OpenAI than generating vibes.

Share

As of this morning, August 18, 2026, Gary Marcus thinks that Sam Altman and OpenAI have just thrown in the towel on capturing the frontier-model lead in the next training cycle round:

Gary Marcus: BREAKING: OpenAI’s Unraveling Has Begun <https://garymarcus.substack.com/p/breaking-openais-unraveling-has-begun>: ‘Their planned IPO is facing headwinds, trust has evaporated, and their burn rate is only getting worse…. Sam Altman’s latest announcement, on Tuesday August 18 (i.e., “earlier today” for those of us on the West Coast), that OpenAI would be pausing, ostensibly for safety reasoning…. As far as I can tell, hardly anyone believed him…. Translation from the user @NIK on X: “We are out of compute”…. Slightly elaborated: Boss Hendricks: “Translation: we need to immediately stop torching cash to provide some semblance of a sustainable business model so we can rush this IPO out the door before the bubble pops”….

The Wall Street Journal’s Berber Jin and Corrie Dribusch just dropped big news: “OpenAI told investors its revenue grew by 18% from the first to the second quarter while its losses deepened… [growing] by $3 billion from q1 to q2, to $12.3 billion. Not a great look given that it added only $1 billion in revenue in the meantime, to $6.7 billion”…. Good news for Anthropic…. And terrible news for OpenAI…

Share DeLong's Grasping Reality Weblog

We had, four months ago:

Nilay Patel: The Ai Industry’s Race for Profits Is Now Existential <https://www.theverge.com/podcast/909042/ai-monetization-cliff-anthropic-openai-profitable-ai-existential-moment>: ‘It’s a make-or-break year for Anthropic and OpenAI, which are facing more pressure than ever to make more cash than they burn…. Hayden Field… senior AI reporter here at The Verge… has been keeping close tabs on both Anthropic and OpenAI…. At some point, the profits have to materialize, or the bubble pops…. You’ve heard me ask some version of this question to scores of CEOs here on this show, and a majority of them have hinted toward the bubble popping — they think some companies will fail in spectacular fashion, some will succeed, and the opportunities, especially the money, are simply too big to ignore. We’re doing this, whether we want to or not — the market depends on it….

AI agents… have radically changed how these companies are thinking about their resources…. Agents are valuable to [code-writing] customers right now, but agents also use far more compute… burning tokens at a rate way faster than these companies anticipated…. OpenAI abruptly killed its video-generation app Sora, ditching a $1 billion Disney licensing deal in the process. Why? It costs too much to run, and OpenAI needs the compute for Codex. We saw it again just last week, when Anthropic decided it would no longer let Claude users burn through compute resources using the OpenClaw agent framework through a standard subscription plan…. The projections these companies have made, which just this week were leaked to the Wall Street Journal, tell a story of mind-boggling growth, to the tune of hundreds of billions in revenue and profitability by the end of the decade. But the most important questions now are can the AI companies pull this off, and what compromises will they make to reach that goal and avoid crashing and burning?…

Give a gift subscription

Following up on his four months earlier prediction: “as nuclear as it gets: OpenAI fails [in 2026]”:

33:31

Get 75% off a group subscription

Even in this context OpenAI’s apparent training pause does surprise me. I had thought that right now OpenAI Codex on the one hand and Claude Code and Cowork on the other were about equal. Some preferred one. Some preferred the other. It seemed largely, these days, a matter of taste and path-dependency. Claude Code wins on code quality, context retention, and multi-agent orchestration; Codex wins on speed, token efficiency, cost-per-task, and fire-and-forget autonomy. Most heavy users run both. And as far as Codex and Claude Cowork are concerned, that race is just beginning, and is too close to call.

And yet it looks like OpenAI’s revenue only (only!) grew from $5.7 to $6.7 billion from the first to the second quarter, while Anthropic’s revenue grew from $4.7 to $11.5 billion. How is this, if Claude Code and Cowork and Codex are rough peers?

Well, first, they were not rough peers at the start of the quarter, on April 1. So perhaps the Q2-Q3 comparison will look very different from the Q1-Q2 comparison.

However, otherwise: The coding tool is a much bigger slice of Anthropic than of OpenAI. Roughly 80% of Anthropic’s revenue is API/enterprise. Anthropic sells agent usage metered by token through its enterprise/API tier, while Codex is delivered inside a ChatGPT Plus/Pro subscription. And OpenAI wants to build loyalty and so does not want to do what Anthropic did to its OpenClaw enthusiasts by cutting off their access. More important, perhaps: Anthropic is the picks-and-shovels supplier to the whole coding-agent ecosystem, not just Claude Code. Claude is the model behind a large share of third-party coding front-ends. with GAAP numbers due in the IPO prospectus.

What I dearly wish to see right now is Anthropic’s S-1 for its forthcoming IPO, which is coming—sometime. FutureSearch “founded in August 2023 by Dan Schwarz… [as] an AI that could predict the future” claims <https://futuresearch.ai/app/p/a/on-what-date-will-anthropic-complete-its-ipo-i> November 4 as the likely IPO date (which means the GAAP financials need to appear in less than a month and a half). Itd further claims:

Three significant overhangs threaten to push the timeline…. Gross-versus-net revenue accounting disputes are highly scrutinized by the SEC; Anthropic reportedly books cloud partnership revenue gross…. Resolving this could require extensive disclosure changes or restatements of ARR metrics. Second, ongoing litigation… over an unprecedented “supply chain risk” designation limits U.S. military contracting and requires sensitive, unresolved risk-factor disclosures . Finally, the sought-after $2T+ valuation demands a staggering $190–200B revenue forecast for 2028…

Leave a comment

My read (which may be very wrong):

1. An Anthropic that comes out of the gate with a $2 trillion market valuation at its IPO is something that does not need the U.S. Defense Department as much as the U.S. Defense Department needs it. Given that standard risk disclosures are simply boilerplate, and FutureSearch is highly likely to be simply wrong here.

2. People who are going to buy Anthropic at the IPO can be easily directed to pro forma financials. The actual GAAP financials will be of relevance only to short sellers who are going to be on the sidelines unless they are stark raving mad given what we have seen over the past decade. FutureSearch is highly likely to be simply wrong here as well: divergence between pro forma and GAAP is also not holding up the IPO.

3. What is, I think, very likely to be holding up the IPO is that a number of organizations that are putting their and their clients’ money on the line here want more than just the second quarter of super-explosive growth from a relatively high base.

Share

Recall that Anthropic’s reported revenue figures are $0.8B for 2025Q2, $4.73B for 2026Q1, and $11.5B for 2026Q2, that those come from leaks or investor-deck slides, and that that is all we know. Other leaks and announcements have been “run rates”:

• $1B ARR as of the start of the 2025.

• $9B ARR as of the end of 2025.

• $65B ARR as of August 1, 2026.

Give a gift subscription

My bet are that these are one-month run-rates at best: i.e., Anthropic booked $5.4B in revenue for the month of July 2026. People will want to see revenue on-track on an S-curve to triple from mid-2026 to moderately late-2028. That means they want to see August and September numbers, that they want those numbers to be good, and that Anthropic is willing to push off the IPO and bet that it can deliver those numbers, rather than scale back the whispered $2T IOP valuation.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##more-signs-of-a-bubble-peak-breaking-news-watch-of-the-day
##macro-outlook
##mamlms
##subturingbradbot
##breaking-news-watch-of-the-day
##openai-the-company-that-had-been-selling-vibes-looks-like-it-is-starting-to-husband-its-cash-in-the-face-of-the-anthropic-challenge-two-peer-coding-agents-and-yet-a-55x-revenue-growthrate-gap
#more-signs-of-a-bubble-peak
#bubble-peak
#ai-bubble
#openai
#anthropic
#coding-agents
#anthropic-ipo

We told you so. Repeatedly. Since 1993 and Newt Gingrich’s takeover of the Republican Party and Bob Dole’s decision that he would rather go along with trying 100% to portray Bill Clinton as a failed president rather than work to make America a better place. (I remember arriving in DC, and finding two weeks later Paul GIgot in the “Wall Street Journal” telling flat-out lies about Treasury Department, that is, my personal, spread of estimates of the economic impact of tax increases.) And we got given a lot of s*** for it. The GOP did not break in 2025. It did not break in 2017. It did not even break in 2012. It broke in 1993:

In this 2012 post, Paul Krugman looks back on more than a decade of being dismissed as “shrill.” His offense was accuracy: he insisted early that George W. Bush was a serial liar pursuing a hard-line agenda, not the blunt honest conservative the commentariat described. He also refused the mandatory pretense of symmetry — the fiction that Democratic caution and Republican radicalism were morally equivalent. By 2012, with a primary field of “not-Romneys” he calls stark raving mad, Krugman argues the party’s condition was decades old. The only new thing was that the pretense of reasonableness had become impossible to sustain:

Share

HOISTED FROM PAUL KRUGMAN’S ARCHIVES: PAUL KRUGMAN (2012): Looking Back With Shrillness

<https://archive.nytimes.com/krugman.blogs.nytimes.com/2012/02/29/looking-back-with-shrillness/>

Brad DeLong notes that the GOP we now see in the primaries has been building for a couple of decades; I can’t help thinking of my own decade-plus in the journalistic trenches.

Early on in my tenure at the Times, I felt I had no choice but to point out the inconvenient truth that the official line of the commentariat was all wrong. George W. Bush was not a nice, blunt, honest guy who happened to be a conservative; he was a serial liar pursuing a hard-line agenda, who among other things deliberately misled America into war.

For this I was labeled “shrill”.

More than that: throughout these past ten-plus years, it has been considered ill-mannered and uncouth, not to mention unacceptably partisan, to suggest that the parties aren’t symmetric — that, for example, the reluctance of Democrats to cut Social Security and Medicare is not equivalent to the GOP’s consistent pursuit of huge unfunded tax cuts, that the occasional desire of Democrats to put evidence in a more favorable light is not equivalent to the constant, raw dishonesty emanating from the right. And pundits in good standing have been expected to make calls for bipartisanship that involve pretending that Republican politicians are actually the kind of statesmen the party used to contain, but no longer does.

So now we see a primary struggle in which the choice is between a series of not-Romneys whose political and policy views are stark raving mad, on one side, and the not-not-Romney who is, maybe, just pretending to share those views. How did that happen?

The answer, as Brad suggests, is that it happened a long time ago. The GOP isn’t just spectacularly unlucky in its menu of candidates; this is what the party has been for decades. Rick Santorum isn’t someone out of left field; he’s always been what you see now, and he was a central figure in his Senate days.

All that has happened now is that the mannerisms have finally gotten to the point that the pretense of a reasonable party is no longer sustainable.

But you weren’t supposed to notice until just about now.

<https://archive.nytimes.com/krugman.blogs.nytimes.com/2012/02/29/looking-back-with-shrillness/>

Brad DeLong here: The thing to understand about Paul Krugman’s 2012 “Looking Back With Shrillness” is that it was not a complaint a much as a diagnosis — and, more than that, a vindication delivered in the flat, tired voice of a man who had been right for a decade, and had gotten little for it but the label “shrill”.

Go back and read what Krugman actually wrote in February 2012. The proximate occasion was a post of mine — “Where Were You in 1993, David Brooks?” — in which I had made the boring, documentable point that the Republican Party then staggering through its clown-car primary had not suddenly gone mad. It had been building toward exactly this for a couple of decades. Paul read that and did what Paul does: he generalized it, sharpened it, and turned it into an indictment not merely of the GOP but of the entire respectable commentariat that had spent ten years insisting the emperor was fully clothed.

His confession — and it reads like one — was that “early on in my tenure at the Times, I felt I had no choice but to point out the inconvenient truth that the official line of the commentariat was all wrong. George W. Bush was not a nice, blunt, honest guy who happened to be a conservative; he was a serial liar pursuing a hard-line agenda, who among other things deliberately misled America into war. For this I was labeled ‘shrill.’”

Now. I have some standing here, because I was a charter member of the order. The Ancient, Hermetic, and Occult Order of the Shrill was not, in its original conception, a club of the angry. It was a club of the accurate.

Give a gift subscription

Read more

Apollo, BlackRock, Blackstone, Brookfield, Goldman, & KKR are about to go all-in taking on equity risk and calling it high-quality debt, apparently in the belief that the money flow to them from the deals is a very large bird in the hand, and that they will find a way to fix it later, if it does indeed need fixing later. Or: private equity now thinks that it is too big to fail, and we know how entities confident that they are too big to fail that are unregulated roll, don’t we?

Last year Jensen Huang said: You need to buy the latest and greatest chips my team has designed right now because the chips we designed that you bought last year are obsolete and effectively worthless. This year, Jensen Huang is saying: You need to lend huge amounts of money with the chips my team designed in the past as security because they are durable income-producing assets with a ten-year economic life.

Share

You see the problem?

That is: Jensen Huang does not remember what his book told him to talk. Or, rather, he is changing books with alarming frequency:

Elisabeth Lopatto: Nvidia’s New Financial Strategy Does Not Compute <https://www.theverge.com/ai-artificial-intelligence/981668/nvidias-goldman-blackrock-gpu-compute-asset>: ‘April – 1805

Napoleon is master of Europe

Only the British fleet stands before him

Compute is now an asset class

I see it is once again time to talk financial innovation. Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR are all working with Nvidia to put together $500 billion in financing to turn compute into an asset class. “This is really the first time that technology chips have become an investable asset class,” Nvidia CEO Jensen Huang said to CNBC. “These are revenue-generating assets now. They’re productive, they’re long-lived, they’re fungible, they’re flexible.” Huang said something very different about Nvidia’s own last-generation Hopper chips last year. “When Blackwell starts shipping in volume, you couldn’t give Hoppers away…. There are circumstances where Hopper is fine. Not many.” So to now be told that chips are actually “revenue-generating assets” that are “long-lived” is… quite frankly, it’s giving me whiplash…. “This is the very beginning, like what it was when I started in the mortgage-backed securities market in the 1970s, and I look upon this as a next future for financial engineering,” said Larry Fink…. Now, for some of you, this may make alarm bells go off…. The AI industry is becoming saturated with data centers, and Chinese open-source models require less compute despite being fairly powerful, both of which seem like potential threats to the notion of ever-growing demand for chips. There is also a far more basic question: Can frontier labs such as Anthropic and OpenAI, which are driving much of the current demand, make money?…

Give a gift subscription

As I keep pointing out to no avail: The gatekeeping question is “Will Amazon, Google, FaceBook, Microsoft, and so on allow Anthropic and OpenAI and anyone else hoping to make a profitable business out of selling LLM-services to make money?” And the likely answer is that they will fight Anthropic and OpenAI tooth-and-nail just as Microsoft made it certain that Netscape would never make money by giving away for free anything and everything that Netscape was trying to sell.

I do not see how you pass that gatekeeping question. If you do pass that gatekeeping question, somehow, then there are more questions:

• What does the world look like in scenarios in which the chip purchasers you, Ms. Private Equity, have financed do not make their amortization payments?

• What does the cash flow from the datacenters and the chips in them that you then have to take over look like in those worlds?

• How do you then explain to your investors whom you promised megaalpha that your major assets are underwater utilities?

I see no scenario in which the fact that the debt is secured is worth anything.

That is, I see compute is now an asset class as let us see if you are foolish enough to make equity investments at a debt-investment price if we simply tell you that it is not equity but debt.

Fight me.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##can-we-sell-equity-financing-of-the-datacenter-buildout-at-a-debt-price-financial-overexuberance-of-the-day
##macro-outlook
##finance
##financial-overexuberance-of-the-day
##jensen-huang-keeps-two-books-last-years-says-old-chips-are-worthless-and-obsolete-while-this-years-says-that-old-chips-are-highquality-tenyear-bonds-elisabeth-lopatto-says-wait-a-minute
#can-we-sell-equity-financing-of-the-datacenter-buildout-at-a-debt-price
#jensen-huang
#compute-asset-class
#private-equity
#ai-bubble
#financial-overexuberance
#hyperscaler-gatekeeping
#underwater-utilities-of-the-future

The bond vigilantes have not pushed the button. They have, however, put a hand on it. Lines have bent upward in Paris, Berlin, London, and Tokyo all at once. That simultaneity is the clue—and it points not at country-level inflation but at the price of trusting governments with your money for three decades. In 1993 James Carville joked that he wanted to come back as the bond market, because then he could intimidate everybody. For decades now things have been on the other side of the hill: people have been desperate to, in real terms, pay governments to keep their money safe. That time is over. At least for now:

The global bond selloff since mid-2020 has been really quite remarkable!

Share

The yield on the 30-year U.S. Treasury bond closed Monday at 5.31%, and touched 5.33% on Tuesday—its highest since the summer of 2007, back before anyone had learned to say “global financial crisis.”

Now run your eye leftward along the line to the trough. You find it in the plague-spring of 2020, at roughly 1%.

That is the whole story in a single picture: the price of long-term money has climbed more than four percentage points—not in a crash, not in a single tantrum, but in a slow, grinding, five-year march that has almost never paused to catch its breath. And widen the lens from the United States to the world, and the line bends the same way everywhere.

• French thirty-year yields sit at their highest since 2008.

• German since 2011.

• British gilts are pushing toward 6%.

• Japanese long bonds are a whisker from their all-time yield record.

Share DeLong’s Grasping Reality Weblog

Whatever is happening here is happening to the entire globe’s stock of safe, long-duration debt simultaneously. Yes, the U.S. is a big player and a Stackelberg leader here. But that simultaneity is, even so, striking.

Leave a comment

Back up: In 1993, watching how fear of rising Treasury yields had laid waste to the spending ambitions of the social-democratic wing of the Clinton Administration he had just done so much to elect, James Carville said:

I used to think that if there was reincarnation… now I would like to come back as the bond market. You can intimidate everybody.”

Give a gift subscription

The line has long outlived its occasion. It names something true. And it names something permanent. Debt markets are the hardest constituency for a government or a political movement to spin, whip, or remove. They turn against you, and you learn the boundaries of your power in a hurry. Since mid-2020, they have been, very slowly, turning.

Ed Yardeni, the man who minted the phrase “bond vigilantes” back in the 1980s Reagan-deficit years thinks the posse is saddling up again. He is writing things like:

We aren’t pushing the panic button. However, we are closely monitoring whether the bond vigilantes might do so….

Get 75% off a group subscription

There is a difference between a market that is annoyed and a market that has decided a government’s fiscal trajectory is no longer credible. Yardeni has watched this movie several times. He is now telling us he is no longer sure which one we will be in come a year from now.

But what is the difference between a normal bond market and “bond vigilantes” becoming active? Interest rates are much higher than they have been for a half-generation, yes. Kenneth Rogoff, however, thinks that it the past half-generation should not in any sense be our baseline. The interest rates of the 2010s and the plague years, he insists, were ultra-low and:

a sharp deviation from historical norm and although they could come back, don’t hold your breath…

Refer a friend

If, looking forward, the supercheap-money decade was the anomaly and not to be repeated, then we are simply living through a return to pre-2007 or even pre-2001 normality. The secular-stagnation world in which the safe real rate sat well below the growth rate was a very real thing. While it lasted. But now the growth rate of the U.S. and world economy and the interest rates on government debt are roughly equal. That means that primary fiscal balance becomes an important goal for governments, which has not been the case since 2001.

So what kind of animal is this selloff?

• Bond-market panic?

• Bond-market discipline based on rational judgments of an unstable and south-trending situation?

• Or simply the end of a period that had seen a highly anomalous global safe-asset shortage, and thus bond market derangement on the rich-pricing side?

The cleanest take I have seen comes from Oxford Economics: “An increased term premium has been the main cause of the jump in US yields,” they conclude. The term premium is simply the extra yield investors demand for the risk of holding a long bond rather than rolling short ones: the price of committing your money for thirty years in a world you cannot forecast. What is crucial in the Oxford view is what is not driving the move: “not primarily an inflation-expectations story.” Inflation breakevens have stayed broadly anchored. The principal drivers of the move have not been higher expected real short rates either. The principal drivers, in their estimation, have been less trust in the dollar and a term premium reconnecting with macroeconomic volatility.

That is: What has risen is the compensation for bond-price uncertainty required for holding an asset that one does not necessarily intend to hold to maturity. In their view, it is not that expected inflation has risen. It is, rather, that not enough people are wishing the government to keep their money safe for the long term. But there is more.

Read more

Plus: Amazon, Facebook, Google, and Microsoft are Agamemnon, Akhilleus, Odysseus, and Nestor, and they are burning cash to make sure no Anthropic, OpenAI, or other Hektor lives to see another sundown…

My personal Visualization of the Cosmic All is that nobody is going to make super-fortunes selling LLM model services to paying customers, whether enterprise or consumer.

Share

Start with the supply side. The problem is this: the moats are shallow.

Nothing stops a rival from offering a near-equivalent model—if only by distilling yours, training a cheaper system on the outputs of the expensive one you spent a fortune to build. Moreover, whatever frontier edge you do manage to open up commoditizes fast: today’s lead is gone in months, as competitors catch up and open-weight models undercut proprietary pricing all the way down toward zero. Put those together. Selling LLM model services looks likely to become a commodity business. And commodity businesses, as every economist since Smith has known, do not earn large economic profits.

This is what I have called the “Airline Scenario”: a technology of huge user surplus and immense usefulness that nonetheless throws off next to no profit for the firms providing it.

The particular cost structure of LLM model-services makes it worse rather than better. Operating costs—compute, electricity, human oversight, error-correction—stay stubbornly high even as prices fall. Moreover, the models themselves offer no escape into economies of scale. They are “train and infer, train and infer”. There is no write-once, run-forever, zero marginal-cost software dynamic here. There is only the perpetual grind of building the next model and then paying again to run it.

Thus when the reported profits do appear, they arrive dressed in “adjusted operating income” and undisclosed math—which is the surest signal that the real profits are not there.

Durable value does exist in this business. Enormous durable value exists. But, in my view at least, it is highly unlikely to sit in the model. It sits in trusted data, in workflow, and in reliability. It sits in the harness and the curated datastores that turn a probabilistic text-generator into something an enterprise, or a consumer can actually depend on. Producer and consumer surplus thus flows to those who control those: end-users, data-cleaners and -curators, and (perhaps) harness providers. And harness-provision looks like it is rapidly becoming a commodity business as well. The raw model, in other words, and almost surely the harness as well, is the cheap and commoditized part. The valuable part is everything wrapped around it.

Give a gift subscription

Moreover, even there is valuable margin harvestable somewhere between the end-user and the model-provider, that margin is under the most aggressive siege since the Trojan War, with Amazon, FaceBook, Google, and Microsoft as Agamemnon, Akhilleus, Odysseus, and Nestor, respectively:

• Amazon’s Agamemnon-like power is infrastructural and logistical (AWS, the cloud, the “pick-and-shovel” business): it is the king who profits by outfitting everyone else’s campaign, with power resting on wealth and command rather than personal excellence in battle.

• FaceBook’s Akhilleus-like persona exerts unbelievable raw force depending on the mood of a single rash personality, one day sulking in his tent, a second going all-in on open source models, and a third day turning academics into multi-billionaires if only they will come work for him and tolerate his adrenaline-rage emotional cycles; my view of FaceBook is that is is, indeed, both devastating and ultimately doomed by its many Akhilleus’s-heel characteristics.

• Google: the wily one with cunning and adaptability, deploying its own models defensively, integrating AI into the browser, out-thinking the threat, and the survivor who will makes it home.

• And Microsoft: perhaps Nestor, because it is also a veteran of an earlier generation of struggles, advancing through through wisdom and alliance-making, advising others when and whom they should fight, not always in their own interests but always in its.

Leave a comment

They are all burning cash in unbelievable amounts precisely to deny everyone else any margin at all. Their spending is defensive. It is to protect the platform monopolies they already hold. That quadruple scorched-earth defense leaves no room for any new platform monopoly to grow up beside them. Anthropic should study Netscape and its fate. And so should those thinking of investing in its IPO.

But there is one problem with this Visualization of the Cosmic All of mine. Back in The Day, I had the same of view of Google as a good business but not a future Leviathan: the only way it could become a future Leviathan, I thought, was by selling its soul to SEO to boost its cash flow and then using that aggressively to scorch the earth around it. And, I was confident, selling your soul to SEO and charging for the eyeballs your second-rate search results trapped into repeated casts of the yarrow sticks would ultimately be self-defeating. Back in The Day, I had the same of view of FaceBook as a good business but not a future Behemoth: the only way it could become a future Leviathan, I thought, was by selling its soul to clickbait rage to boost its cash flow and then using that aggressively to scorch the earth around it. And, I was confident, selling your soul to clickbait-rage and charging for the eyeballs your outrage-bait feeds had trapped into doomscrolling would ultimately be self-defeating.

Guess what? Selling your soul to SEO was not self-defeating. Neither was becoming the master necromancer of rage-bait doomscrolling.

Am I missing something similar here and now?

Get 75% off a group subscription

And so we have:

(VERY PARTIAL-) CROSSPOST: GARY MARCUS: The hyping of Anthropic’s IPO

<https://garymarcus.substack.com/p/the-hyping-of-anthropics-ipo> <http://garymarcus.substack.com>

Some strong claims, dissected

Gary Marcus

Aug 16, 2026 ∙ Paid

I am watching something fascinating unfolding in real time. Anthropic is in an SEC-monitored “quiet period” before its IPO (expected sometime in the fall), somewhat limiting its communications (though not entirely; Amodei discussed governance matters on X yesterday, a rare appearance there for him, and the company continues to report on safety issues and so on). But that hasn’t stopped leakers and generative AI bulls from trying to convince audiences that Anthropic’s finances are spectacular.

Aside from the Friday Reuters report above in which Anthropic is said to be “projecting 2028 ‌revenue of roughly $190 billion to $200 billion”, based of course on undisclosed math, “according to two [unnamed] people familiar with the company’s financials”, we also heard yesterday about how amazing their Q2 was (further details below), by way of documents that were leaked to Bloomberg News. Anthropic may be “quiet”, but its fans are not.

Moreover, in a Friday All-In podcast interview, Gavin Baker claimed that Anthropic was making money on every token. A couple weeks earlier the prominent podcaster Dwarkesh Patel projected (in a claim he partly walked back, see below) that “Anthropic likely ends the year with ~$100-150B of revenue”

Below, I dissect these claims and explain what’s been left out…

<https://garymarcus.substack.com/p/the-hyping-of-anthropics-ipo> <http://garymarcus.substack.com>

Brad DeLong here: That is all that Gary Marcus puts above his paywall.

However, what is below Gary Marcus’s paywall is free with a seven-day trial.

I do recommend that you do so to read the whole thing and to at least try to figure out whether Gary Marcus is someone you might well want to pay for.

Here is my summarization of his points:

Marcus believes that the bullish financial claims circulating about Anthropic ahead of its fall 2026 IPO:

• a blowout Q2,

• $100–150B annual revenue by 2026-end, and

• now profitable on every token;

Refer a friend

are either

• misreading,

• cherry-picked from an anomalous quarter, or

• unverifiable

Share DeLong's Grasping Reality Weblog

In Marcus’s view, the real question remains unanswered:

• Can Anthropic ever be durably profitable once the full cost of constantly retraining rapidly-obsolescing models is counted?

• For if Anthropic cannot, nobody else can.

Refer a friend

Marcus believes that the second quarter of 2026 was an anomaly:

• It was the peak of a now-dead “tokenmaxxxing” fad.

• It was before Chinese open-weight models arrived to substitute for (the very impressive) Claude harness.

• It was before the price wars produced by Chinese-lab entrance and serious attempts by other US labs to gain market share by underpricing Anthropic’s Claude.

• It was when Anthropic’s revenue was substantially inflated by a unique demand from SpaceXAI.

Give a gift subscription

Thus quadrupling that quarter’s revenue to give Anthropic a $50 billion current run rate overstates the trend. And to get to Dwarkesh Patel’s $100–150B year-2026 revenue for Anthropic requires a further explosion of demand for Claude harnesses without a fall in price, or in a super-explosion of demand with expected price declines. Moreover, Marcus argues, revenue is not equal to profit in an environment in which not only electricity and amortization are major variable costs but in which training new models approaches being a variable cost as well. Gary Marcus:

I don’t know much about [Gavin] Baker or his motivations or his calculations…. [My] most charitable read… is that… (a) the largest costs are training and researching models rather than running them, and that (b) in the moment of execution… [they] actually make money. That would be fine, if the half-life of the models was like the half-life of train tracks. But it’s not…. Models become old news in a matter of months. Anthropic’s market niche… is… training top models that quickly lose their edge. [That] is thus an ongoing cost of business for them, which can’t be ignored.

My guess is that once that massive cost and its depreciation is factored in, Anthropic is in fact not making a profit on every token. But again we just don’t have much transparency around these things. Hearsay on All In is not enough to clear things up…

Get 75% off a group subscription

For Anthropic’s ongoing new-model training costs to have a first-order effect on its profitability, it has to be the case that the models continue to get substantially better with every passing month, rather than models being “good enough” at natural-language processing and big-data high-dimension flexible-function classification so that usefulness comes from trusted and curated datastores and the harness though which language outputs from LLMs are then fed through deterministic software pipelines to emerge on the other side with trusted ground-truth outputs that can be then fed back into LLMs again. And if models are getting that much better every month, then they are either (a) adding more potential value to users, hence a provider oligopoly will be likely to be able to charge more to high-value users, or (b) selling to rapidly expanding market for whom new use cases become feasible, in which case you can make it up on volume.

Thus I find myself wondering: Is moving model-training costs to the “variable cost” bucket really the win that Gary Marcus thinks it is going to be for his bear thesis?

I mean, we have here:

• Asymmetric skepticism: Marcus “doesn’t doubt” the revenue numbers but heavily discounts the profit ones. That is not implausible. But his priors as a prominent generative-AI skeptic shape which claims get scrutiny.

• Hearsay cuts both ways I: Marcus rightly dismisses “hearsay on All-In,” which has been a striking source of misinformation: Trust All-In, and you are out there believing Travis Kalanick, who claimed his conversations with LLMs were at “the edge of what’s known in quantum physics…. I’m doing… vibe physics…. I’ve gotten pretty damn close to some interesting breakthroughs just doing that...” with all of the podcast hosts gullibly swallowing it.

• Hearsay cuts both ways II: Note that several of Marcus’s own load-bearing facts, like the importance of SpaceX subsidies, the final death of tokenmaxxxing, and switching to China-provided models already in train are also it the realm of impressionistic near-hearsay.

• The “disco fallacy” is a frame, not evidence: Whether Q2 was a transient peak or an early point on a steep adoption curve is exactly what is now unknown.

• The model-depreciation argument needs numbers: The train-track analogy is rhetorically strong but unquantified — if inference volume scales fast enough, per-token margin can exceed amortized training cost even with short model half-lives. Marcus offers a “guess,” not a model.

• The commodity-business assumption: Assumed, not argued. (echoed in the top comment) is assumed, not argued.

Do not get me wrong: I believe the commodity-business assumption. But it is an assumption, not something demonstrated.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##very-partial-crosspost-gary-marcus-is-anthropic-making-money-how-much-money
##macro-outlook
##mamlms
##subturingbradbot
##crosspost
##nobody-ought-to-get-superrich-selling-llm-services-anthropics-ipo-and-the-airline-scenario-the-problem-is-that-back-in-the-day-i-was-the-same-bear-on-google-and-facebook-and-i-was-very-wrong
#anthropic-ipo
#gary-marcus
#llm-economics
#platform-monopoly-scorched-earth

Russia’s refineries are substantially dark; Saudi’s Jazan has been hit, how badl I do not know; and the Strait of Hormuz has been near-shut since February. The crude market: it notices, but not that much. However, there is serious pain downstream, in the diesel crack spreads. Crude oil is not that unplentiful; the machines that turn it into diesel are not:

We have, today:

Share

From:

No. 1: Weekend Thoughts <https://no01.substack.com/p/weekend-thoughts-bd0?> ‘Diesel crack[ed product to crude-oil price spread] at a record [$]98.169[/barrel]…

Give a gift subscription

Because bigger molecules pack more tightly and because cracking crude oil adds hydrogen from outside to the product, we get 1.05 barrels of product out of one barrel of crude oil input. Of that 1.05 barrels of product, in US refineries typically get: 0.30 barrels of diesel, 0.50 of gasoline, 0.10 of jet fuel/kerosene, and 0.15 of other products like propane, feedstock for plastics and such, asphalt, residual fuel oil, and so forth.

The diesel and the jet fuel/kerosene crack spreads move almost in lockstep.

This is a genuine crisis in the petroleum markets. But it is not so much a crude-supply crisis as a refining crisis. The world has lost roughly 7.5 million b/d of refining capacity with Russia effectively halting diesel exports as perhaps 2/5 of its capacity to refine crude oil is off line, with the Houthi strike hitting Saudi Aramco’s Jazan refinery, with the near-closure of the Strait of Hormuz since February.

Right now the diesel spread is roughly $100/barrel, gasoline is $35 (recreational and discretionary passenger-car driving is down because of its higher demand elasticity), jet fuel/kerosene is $50 (flight numbers are also down), and residual is $5.

Residual’s crack spread is usually negative: refineries wish that they could turn it into something else, but they cannot. Moreover, the 2020 ban on excess sulfur emissions is effective, and is effectively a ban on cheaply burning any residual from high-sulfur crude oil. So it is hard to generate an excess demand for the stuff.

Note that this crack-spread crisis is not in anticipation of the expiration of the “ceasefire” date coming to pass. It is on top of whatever a full closure of Hormuz would do to the crude-oil price.

But, admittedly, that might not be much: there has been very little traffic through the Strait of Hormuz since February, or so we think, compared to the 130 ships per day that passed through the Strait of Hormuz in the Before Times:

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

Leave a comment

##again-more-of-a-refining-crunch-than-a-crude-oil-supply-crunch-chart-of-the-day
##chart-of-the-day
##neofascism
##macro-outlook
##lose-75-million-barrels-a-day-of-refining-capacity-along-with-crude-oil-supply-and-the-shortage-shows-up-more-in-the-refined-product-prices-than-in-the-crude-oil-price
#again-more-of-a-refining-crunch-than-a-crude-oil-supply-crunch
#refining-crunch
#strait-of-hormuz
#jazan-refinery
#russia-diesel
#supply-shock

The striking disjunction between what’s on the label vs. what’s in the can with respect to the Kevin Warsh communications story. It turns out the new Fed Chair does want to guide the market, just not on the record. Which leaves the rest of us to guess which “people familiar with his thinking” are actually familiar with his thinking:

Caudia Sahm says, rightly:

Claudia Sahm: <https://x.com/EconBerger/status/2085352416200003738>: ‘Warsh is the Fed Chair. We should all be familiar with his thiking. We should not have to rely on anonymous sourcing…

Share

This is à propos of:

Claire Jones & Kate Duguid: Kevin Warsh to Stick with Lean Fed Messaging Despite Market Backlash <https://www.ft.com/content/debe096f-ec89-424f-a8ca-d3843ef53549?syn-25a6b1a6=1>: ‘People close to Warsh said he acknowledged that he had made mistakes in his first 10 weeks…. People close to Warsh noted that market-based measures of inflation expectations remain low and have fallen back in recent days…. President Trump’s repeated attacks on… Powell…have… complicated the new Fed chair’s efforts to convince investors that he is a committed inflation fighter. Warsh would be prepared to raise interest rates at September’s meeting if inflation readings released in coming weeks are hot, and markets ratchet up their expectations for increases in borrowing costs, the people familiar with his thinking said…. Interest rates remain the primary tool [of monetary policy] for now — and would be used at forthcoming meetings if needed, the people added…

Share DeLong's Grasping Reality Weblog

First, one has to strike a note of amazement at the most extraordinary self-undermining of a headline and a lead by a story: sending people “familiar with your thinking” and “close to you” out to talk to the press about your thinking is neither “lean Fed messaging” as the headline says, not is it a “stripped-back communications style”, as the lead sentence claims.

One has to imagine someone, either the headline writers or the reporters, or maybe all of them, chortling together at the disjunction between what is on the label and what is in the can with respect to this story.

Give a gift subscription

And we have, added:

Guy Berger: ‘It’s good that Warsh is abandoning his opposition to (Delphic) forward guidance and talking to the press, though it would be better if he also did it at the press conference…

Get 75% off a group subscription

So it turns out that Kevin Warsh is not opposed to forward guidance being carried out by “people familiar with his thinking”; he is only opposed to attaching his name to it, I think because attaching his name to it would then anger either the markets or Trump, while if it is done through others he has some plausible deniability when either markets or Trump come after him. But having to guess which of the people claiming to be “familiar with his thinking” is actually, you know, familiar with his thinking may be the worst of the nearby timelines.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##the-fed-chair-who-whispers-kevin-warshs-lean-messaging-is-anything-but-laugh-of-the-day
##macro-outlook
##neofascism
##laugh-of-the-day
##forward-guidance-but-outsourced-to-anonymous-sources-familiar-with-the-thinking-of-close-to-the-new-fed-chair-kevin-warsh-it-is-funny-one-has-to-laugh-also-cry
#the-fed-chair-who-whispers-kevin-warshs-lean-messaging-is-anything-but
#kevin-warsh
#forward-guidance
#lean-messaging
#monetary-policy
#federal-reserve
#delphic-guidance

We used to think the sacrifice victims of the Shang Dynasty were prisoners sacrificed within days of reaching the capital. The isotopes locked in their bones tell a grimmer story: years of forced labor, then execution. For centuries, the captives called qiang were thought to be a single enemy people. The truth is worse: sheep-people 羌 qiang wasn’t an ethnicity, it was a status — anyone the Shang caught and marked for death:

“Sheep=people”. People who herd sheep? Or people whom we treat like sheep?

Patrick Wyman is one of my favorite podcasters. His book, The Verge: Reformation, Renaissance, & Forty Years that Shook the World <https://www.hachettebookgroup.com/titles/patrick-wyman/the-verge/9781538701171/>, is just about perfectly pitched for my undergraduate students. Moreover, it is a perfect capstone to the argument, which I buy, made by Patricia Crone, that it was the incompetence of Europe’s predatory élites at collective action to successfully run their version of an agrarian-age society-of-domination that opened the way to the cultural-economic-industrial explosion we have seen over the past half-millennium.

Share

Here we have Patrick Wyman diving into the extraordinary culture of the extraordinary society of domination that was China’s Shang Dynasty 商朝 from -1600 to -1050

Give a gift subscription

The Shang state appears to have run on an extraordinary self-reinforcing engine of war and sacrifice:

• the army captured people at the periphery,

• they were then used as forced labor at the center,

• after which they were killed to please?, to appease?, to somehow feed? a bizarre conglomeration of ancestral spirits,

• with sixteen sacrificed in the grave of Queen-General Hao 妇好,

• with the largest mass sacrifice we have found amounting to some 350 people;

• not a proto-bureaucratic empire of the kind we are used to from the Middle East;

• rather, a sacred and sacrificing order of competing lineages commanding soliders and slaves, where the the scale of the human sacrifice your lineage could undertake was the principal way to gain lineage prestige via spectacle.

Get 75% off a group subscription

Let me stress scale. The population of the Shang capital of Anyang probably peaked in the low six figures, in an age where the entire population of China was perhaps 8 million.

Now we do see in early agrarian age China, all the way back to the year -4000, patterns that were to come to dominate in the Shang: ancestor worship, human sacrifice, and divination. The Shang Dynasty did not originate these. But it did undertake them at industrial scale.

The Shang appear to have run among the very worst of the many bad societies-of-domination we see over the long agrarian age.

Refer a friend

Perspectives: Past, Present, and Future, by Patrick Wyman

The biggest ideas shaping our world - its past, present, and future.

CROSSPOST: PATRICK WYMAN: The Human Sacrifices of the Shang Dynasty

<https://patrickwyman.substack.com/p/the-human-sacrifices-of-the-shang> <https://patrickwyman.substack.com/>

Perspectives: Past, Present, and Future, by Patrick Wyman

The Human Sacrifices of the Shang Dynasty

Mei kept his head high as he ascended the steps of the platform, eyes forward and not on his bare feet scraping the compacted earth. The guard walking behind him didn’t have to use the sharp point of his polearm to prod Mei along. Despite knowing what awaited him at the top of the steps, he went there of his own volition, if not altogether willingly…

Read more

11 hours ago · 5 likes · Patrick Wyman

Patrick Wyman

Aug 13, 2026

Mei kept his head high as he ascended the steps of the platform, eyes forward and not on his bare feet scraping the compacted earth. The guard walking behind him didn’t have to use the sharp point of his polearm to prod Mei along. Despite knowing what awaited him at the top of the steps, he went there of his own volition, if not altogether willingly.

The ropes binding Mei’s wrists chafed at the old scars underneath. They’d healed poorly and bled constantly, layer after layer of healed tissue flaying open at the slightest friction. The joints themselves were weak and painfully thin. Back home in his village, Mei had been a strapping man, fond of (and good at) wrestling with his brothers and cousins; he was often chided for scooping more than his share of millet porridge out of the communal pot. His large size had made him one of the first choices to defend the village from raiders or to help rustle cattle from the pastoralists who lived in the dry grasslands beyond the hills.

Thanks for reading Perspectives: Past, Present, and Future, by Patrick Wyman! Subscribe for free to receive new posts and support my work.

Listen here and be sure to subscribe to Past Lives:

Mei supposed that made him an important man. His people, across villages up and down the valley, did look to him for leadership at times. Somebody had to do the rough jobs like leading the lads out to fight, and Mei didn’t mind if it meant having a bigger house and a few slaves to work in it. To his Shang captors, that made Mei a chief of sorts, a captive to be prized. These men hadn’t even taken him in battle, worst of all: He had been out in the fields, weeding and shooing away birds from the ripening millet, when the soldiers surrounded the village.

Now, after the months spent walking east through the mountains and the months spent here in the bronze foundry, there was little left on Mei’s bones but skin. Even that was peppered with wounds, lesions, burns, and calluses, his body a ruin from the blisters on his feet to the whipweals on his back.

This city was enormous, bigger than anywhere Mei could have ever imagined, filled with more people than he thought lived in the entire world. Their faces would be the last ones he saw in life. Weak, wounded, and alone, Mei knew there was no point in running now: With a guard behind him and two more at the top of the steps, plus the waiting executioner with his wide-bladed bronze axe, he knew he was doomed to die.

Still, Mei kept his head high, his steps assured. His captors could take his life, as they had taken his freedom and his labor, but he would choose the manner in which he left this world.

Break

Hello, my friends, and welcome to another episode of Past Lives. I’m your host, Patrick Wyman. Thanks so much for being here today.

The Shang Dynasty marked the beginning of recorded history in China. For the first time, we have contemporary written texts that can add depth, detail, and flavor to the mute testimonies of the archaeological record. There had been cities, states, and kings in China long before the Shang, going back deep into the Neolithic period more than a millennium earlier, but the 13th century BC marked the beginning of an unbroken literacy and an ideal of centralized rule, both of which have endured through the years to the present. The script used to inscribe the Shang oracle bones with records of divinations and sacrifices is the obvious ancestor of the characters that remain in use today, the language an early form of Chinese. The Shang ruled the largest state that had ever existed in East Asia, and set the template for every subsequent dynasty and ruler. In ways both large and small, the Shang are the beginning of China as we have come to understand it.

In our last episode, we discussed the Shang world from above, through the tomb of queen and general Fu Hao. You’ll remember the thousands of objects made from bone, jade, and bronze that were placed in the burial chamber with her: metal vessels used in elaborate ancestor-worship rituals, turtle shells recording past divinations, jades so old they were already antique, treasures far beyond what the average resident of the vast capital city of Anyang ever would have seen or held during their lives. The most telling inclusions in Fu Hao’s tomb, however, were the sixteen people who were executed and interred with her. Human sacrifice was a shockingly common practice at Anyang, and the Shang world revolved around a sacrificial economy in which prisoners taken in campaigns were brought back to the capital to perform hard labor until their eventual executions.

Today, we’ll turn our attention to the victims of this practice rather than the perpetrators. Thanks to reevaluations of the oracle bone inscriptions and cutting-edge archaeological science, we now have some idea of who these people were, where they came from, and what their lives were like. Shang political life and the power of the state depended on the military campaigns in which people were taken and enslaved. Their toil made the lavish lifestyles of Anyang’s elite possible. Their deaths were the currency that paid the spiritual tolls demanded by a baffling array of ancestral spirits. But these people weren’t curre ncy, or labor, or captives; they were people, with families, experiences, hopes, and dreams. The Shang deliberately and systematically destroyed their lives. The least we can do more than 3,000 years later is return them to the center of the story.

The Shang knew their capital city as Dayi Shang, “Great Settlement Shang,” and between its founding in 1300 BC and its abandonment 250 years later, it was almost certainly the largest, most populous urban site on the planet. It was a sprawling, rather loose city compared to its contemporaries, places like Pi-Ramesses in Egypt or Babylon in Mesopotamia, but it was much bigger: At its peak, Anyang covered nearly 14 square miles, 3600 hectares, with an estimated 100,000 residents. Even those numbers might be low, because Anyang wasn’t bounded by a circuit of city walls; it simply grew less and less dense at the edges, petering off into the surrounding pastures and fields of wheat and millet. There were certainly walls at Anyang - huge, thick structures made of rammed earth - but they enclosed individual districts of the city, namely those of the nobility and the royalty who ruled the Shang world. Anyang was essentially a collection of communities that aggregated into mile after mile of low-density sprawl: clusters of workshops surrounded by the homes of the laborers, many of whom were enslaved; walled compounds the size of a Neolithic city-state containing individual noble (but not royal) lineages and their hundreds or thousands of subordinates; and at the center, bigger and grander than the rest, the fortified palaces, gathering-halls, and tombs belonging to the Shang kings themselves.

Everything at Anyang was simply bigger, vastly greater in scale, than anything that had come before in Chinese history. The capital city was larger, of course - twice the size of the previous center of power at Zhengzhou - but so were the mountains of slag produced by industrial bronze-casting operations in the city. In fact, the largest known bronze artifact from anywhere in the ancient world was produced at Anyang: The Houmuwu ding - a three-legged bronze vessel - weighs 830 kilograms, more than 1800 pounds. That ding was among the copious treasures that accompanied Shang nobles and royalty into their graves. Fu Hao’s lavish tomb contained more than 1,000 individual artifacts, many of which had been purposely carved or cast for the occasion; that was a fraction of what a Shang king’s burial chamber might contain. Moreover, despite her military role, Fu Hao wasn’t buried with a chariot or horses, common accoutrements for the Shang elite. And while sixteen individuals lost their lives as human sacrifices to accompany Fu Hao into death, that too was a minuscule number compared to the blood spilled on many other occasions. If we add up all the people marked for death as sacrifices in the oracle bone texts discovered at Anyang, we arrive at a figure a bit north of 13,000. Since the texts we have are only a small portion of what once existed, and not all human sacrifices were recorded with divinations, the true number is surely far higher. Archaeologists have consistently found pits or groups of pits containing the remains of fifty or more victims. The largest concentration yet recorded involved 339 executions, all deposited in a single act of mass sacrifice. The textual and material evidence align: In a little over two centuries, thousands of people were killed at Anyang.

Why did the Shang do this? What was the logic behind sacrificing dozens or hundreds of captives at once? To answer these questions, we have to understand how the Shang saw the world. It would be easy to think of them as the forerunners of the bureaucratized, centralized empires of later Chinese history, with coteries of administrators, a standing army, and regular procedures for controlling a vast territory; but that’s not actually how the Shang state functioned or what its rulers valued. Its kings weren’t bureaucrats or cosseted emperors; they were equally spiritual and political animals, just as comfortable at a divination or presiding over a sacrifice as they were on a hunt or at a feast with their nobles. Even for more recent periods - the European Middle Ages, for example - we often overemphasize the pragmatic sides of kingship, like making beneficial marriages and starting wars, at the expense of the sacral aspects. The prevailing idea of a healing touch, or the king’s mythical two bodies - one physical, one eternal - indicates that there was more happening than just practical rulership. The further back we go in time and the wider we cast our geographic net for comparisons, the stranger kingship seems to become. We’ve talked at great length about the centrality of the king in ancient Egypt, the semi-divine aspects of his person and rule; the Shang, by comparison, took that concept in directions that seem even stranger to a modern eye.

At heart, the Shang didn’t imagine their state or the broader society around it as a singular entity. Instead, it was a collection of distinct lineages stretching back through time to honored ancestors. After death, these ancestors weren’t gone from the world; they remained active in the lives of their descendants, and it was incumbent on the living, particularly the most important among them, to provide sufficient sacrifice to their ancestral spirits. For the members of the Shang dynasty and their contemporaries in the Central Plains, spirits of all kinds were a dangerous and omnipresent force in their lives, causing all manner of calamities. A flood? There was a spirit to blame. A failed military campaign? An ancestor was angry about something. An imminent harvest? Best to avoid an ancestral spirit’s unhelpful interference by appeasing them with a sacrifice in advance. Everyone had ancestral spirits; the more elevated a lineage, the more important its ancestral spirits, and the Shang had the most illustrious of all. It was therefore their spirits who required the greatest and most consistent appeasement. When married to a state apparatus that could mobilize thousands of soldiers and control much of what is now present-day China, this became the basic logic underlying industrial-scale divination and sacrifice. I love the way the archaeologist Elizabeth Childs-Johnson sums up this dynamic, as “the fusion of ‘shamanic’ belief with the institution of rulership and its power mechanisms.”

The Shang didn’t invent ancestor worship. There are signs of this belief system dating back nearly 3,000 years before, in the archaeological cultures that eventually developed into the Bronze Age societies of the Central Plains. At the Neolithic village of Jiangzhai, dating to around 4000 BC, different residential portions of the settlement were oriented toward different cemeteries. Archaeologists have interpreted these segments as representing the living and dead members of lineage groups that persisted through time. The Shang didn’t invent sacrifice, even human sacrifice, as a mechanism of ancestral appeasement: Sacrifice is simply the deposition of something valuable as a gift to a spirit. In China, human life was a valued currency in this exchange dating well into the third or even fourth millennium BC. Divination too is far older than the Shang, with instances of drilled and cracked bones dating back centuries before them.

What changed with the Shang, in all of these aspects, was scale. Lineages seem to have been the fundamental building block of societies in China dating back deep into the Neolithic, but the competition between and accumulation of resources by these kin-groups also turned them into the building blocks of the Shang state. In fact, we should probably see the vastly increasing scale of sacrifice - dozens or hundreds of captives instead of a bare few buried underneath a new palace - as evidence of how powerful the kin-groups had become; they could afford to expend hundreds of cattle and people in rituals designed to glorify their ancestral spirits. Lesser lineages might sacrifice one or a few captives at the death of a valued family member, but that could never suffice for Shang royals. If their ancestors required such appeasement, then those who claimed descent from them must be equally prestigious and potent in the present.

This logic became the driving force behind the Shang state as a whole. Violence underpinned it all: Violence at the margins, where neighboring groups would be raided for captives and plunder or brought under the Shang yoke, and violence at the center, where the lives of those captives were expended to further the prestige of the Shang. As a much later text from the Spring and Autumn period, the Zuo Zhuan, puts it, “the great affairs of the polity are sacrifice and war.” More than that, it was a self-reinforcing cycle, and it peaked at precisely the time Fu Hao was buried with sixteen human sacrifices.

The Shang had a specific word for the captives they brought back to Anyang after a successful military campaign: qiang. For a very long time, scholars working on the oracle bone texts thought qiang was an ethnic designation, a name for one of the neighboring groups living at the edges of the Shang state. The truth is rather more prosaic and depressing. What made someone fall under the label qiang wasn’t their specific ethnic identity but the fact that they were a commodity destined for expenditure in the sacrificial market. One oracle bone inscription, for example, records the capture of a leader named Mei, plus 24 other people - presumably military-aged men - along with 100 captive women, 1,570 scalps, and assorted sets of weaponry and armor. Other inscriptions list the captives, labeled as qiang, alongside livestock. These same oracle bones also often record the specific ancestors to whom the cap tives were sacrificed: Mei was killed to appease Ancestor Ding; the 100 captive women were killed on behalf of a certain Ancestor Yi. Occasionally bits of human skull were used as oracle bones, but most captives were simply dumped into pits in groups of ten after their executions. Mei’s head was displayed as a trophy. Sacrificial deaths were gruesome: decapitation and dismemberment were standard, but other captives were cut in half at the waist or even buried alive. Through this activity, the king domesticated the people of enemy groups, reducing human beings to the same level as the livestock who were sacrificed alongside them. He was powerful, and they were weak. It was a brutal logic reinforced through systematized death.

And yet the people whose bodies were tossed into those pits ringing the burials of the Shang elite were human beings. They had lives before their capture and the long march to Anyang. They had lives at Anyang, sometimes for years, prior to their executions. Now, thanks to recent research, we can reconstruct their backgrounds with more depth and detail, and hopefully restore some of the humanity the Shang and their allies stole from them more than 3,000 years ago.

Our first order of business is to establish where, exactly, the people sacrificed at Anyang - and the non-elite inhabitants of the city more generally - came from. While the oracle bone texts suggest that sacrificial victims in particular were drawn from captives, the qiang label applied to them obscures their geographic and cultural origins. The scholars working on the cutting edge of this field would say that was an intentional act of dehumanization: Even in death, millennia later, we’re still trying to work past the Shang worldview to the underlying reality of who these people were. Luckily for us, recent studies have enabled us to do precisely that. If you think back to previous episodes in this season, you might remember stable isotope analysis. As living creatures grow, we absorb the distinct local isotopic signatures of the food we eat and water we drink, which are locked into our bones. We can then sample human bones and analyze it to determine those isotope values, the most commonly used of which are oxygen, strontium, nitrogen, and occasionally sulfur. These values tell us about geographic origins, since different regions have different background levels of isotopes, and also about the composition of an individual’s diet. Different bones lock in the isotopic signature from different times in our lives: Our tooth enamel is more or less set by adolescence, so the isotope values there tell us about childhood; the cortical bone of the ribs, by contrast, is constantly remodeling over the course of our lives, so values from that part of the body tell us about a person’s last years before death.

This is particularly useful for the study of the Anyang captives, because - assuming we have bone samples from different sites on the body - we can compare the values from different times in their lives. If the sacrificial victims were born near Anyang, then the isotopic ratios in their tooth enamel should match local values scientists calculate from the landscape. This may well have been the case for the high-status retainers who sometimes accompanied their kings and nobles into death, a type of sacrifice known as renxun, as opposed to the simple “human offerings,” renshen. If they came from elsewhere, however, those values will be different. If the values from their cortical bone match Anyang but their tooth enamel gives us a different number, then we’ll know that they came from elsewhere but lived at Anyang for some time before they died. If we’re lucky, then we might be able to match the differing values to other regions near the Shang heartland and provide a precise origin.

Archaeologist Christina Cheung, whom I had the pleasure of interviewing a few years ago on Tides of History - season 4, episode 85, if you want to check it out - has carried out extensive bioarchaeological research on the human victims from Anyang. One study she led examined 68 individuals who had been sacrificed. Some were represented only by portions of their bodies, specifically the skulls, which had been dumped in a pit near a royal burial, while other sets of remains lacked the skull altogether, for fairly obvious reasons; others were relatively intact. As a control, Cheung and her colleagues also examined 39 local residents of Anyang, commoners who had been buried in humble but normal graves, to compare the isotopic values.

The results were striking. Practically all of the sacrificial victims had isotopic values that pointed to origins far from Anyang; only one person who was sacrificed was determined to have been a probable local. This lone individual was almost certainly one of the more elevated retainer burials, renxun, rather than a brutal dismemberment of a human offering, renshen; the local sacrifice had also consumed a great deal more animal protein over the course of his life than the common victims, a likely sign of higher social status. The non-locals had eaten a slightly different diet: Both they and Anyang residents ate mostly millet, but Anyang natives supplemented that fare with wheat, rice, or local wildlife. Cheung’s most significant finding, however, is that the sacrificial victims’ diets had changed in the years immediately before their deaths. The most convincing explanation for this phenomenon is that prisoners of war, those labeled qiang in the oracle bone texts, weren’t killed immediately upon arrival in the Shang capital. Most scholars had previously believed that captives were marched to sacrificial platforms within days, but the isotopic evidence demonstrates beyond doubt that these victims had been kept alive and working for years before their deaths. We know that because bone from elsewhere in the body has different values than their tooth enamel, meaning that they lived long enough in Anyang to carry its distinct isotopic signature, including the local diet.

We can be more specific about the victims’ geographic origins than just saying they weren’t from Anyang. The bulk of the isotopic data suggests that they came from somewhere to the west, particularly the northwest. While the Shang dominated the Central Plains, they and their political allies and cultural relations were far from the only groups living in present-day China. It was a diverse place, with varying lifestyles, languages, material traditions, and genetic ancestries. Some of these groups were pastoralists, links to the dynamic world of the Eurasian steppe and the probable conduit of such innovations as domesticated horses and cattle, chariots, wheat, and not least, bronze-working itself. Scholars generally assumed that the Shang had a particular dislike for pastoralists and targeted them in their incessant wars and raids, but the isotopic data doesn’t support this conclusion; the people Cheung and her colleagues studied had all relied on agricultural products prior to their capture, rather than the dairy and meat common among pastoralist groups. There is variation among the sacrificial victims, however, suggesting that they weren’t necessarily members of the same cultural or ethnic group. That is revealing, because it reinforces the argument that the Shang didn’t particularly care who they were raiding or sacrificing. Anybody could be qiang; it was the fact of their capture and transportation to Anyang that made them so, not any specific origin.

Once they arrived at Anyang, some - but not all - captives were immediately sacrificed: perhaps in thanks for a victorious campaign, or to sate the anger of a spirit as revealed in a recent divination. There is some uncertainty as to where the captives lived in Anyang, and under whose control: They might have been held in prisons controlled by the royal dynasty, or kept in the compounds of the noble lineages that towered over the commoners’ districts. Wherever they lived, it’s certain that they were put to work. The Shang weren’t ones to let labor sit around unutilized, and there was nothing sacred about the human beings or livestock destined for sacrifice. Life at Anyang was hard for everyone who wasn’t among the ruling elite: Studies show extraordinarily high rates, over 80 percent, of dental enamel hypoplasia among the residents, indicating astronomical levels of childhood stress. Forty percent of adults had osteoperiostitis, a generalized stress reaction visible in the skeleton. For women, those general stresses and workloads continued throughout their lives; adult men were more likely to show signs of backbreaking labor, particularly in the upper body. To my knowledge, these paleopathological studies haven’t been carried out on sacrificial victims, but there’s no reason to think they labored under better conditions after their arrival at Anyang. If anything, they were probably treated worse, and worked harder, before their executions.

Most of the sacrificial victims we know from the archaeological record were adolescent and young adult males, aged between about 15 and 35 at the time of death. Oracle bone inscriptions tell us that female captives were sometimes killed as well, but excavations have yet to find them in significant numbers. It seems that adult men were preferred as sacrifices, though exactly why is unclear: They may have had the most value to the spirits, or perhaps it was considered unwise to keep large numbers of military-age men hanging around the capital in captivity. If the general point of the executions was dehumanization and dominance, however, then picking men of fighting age was the strongest way to demonstrate strength: Watch as we turn men who were once warriors or who could have been warriors, and thus a potential threat to us, into livestock under the blade. Watch as we throw their bodies into pits, denying them funeral rites and the offerings of their descendants. See, from that, how strong we are.

We don’t have to accept the Shang framing of these people. They came from agricultural villages and pastoralists’ camps. They fought among themselves, and surely raided lands the Shang considered theirs, but mostly they farmed and herded, made pottery and tools, grew up and raised families. Then, one day, the Shang armies arrived on their doorstep, led by highly competent generals like Fu Hao. Perhaps the men were taken in battle, but it’s more likely that the Shang simply swept up whomever they found. The survivors witnessed the deaths of friends and kin, the taking of scalps and heads, and all the other diligently recorded atrocities we find in the oracle bone inscriptions. Such was the scale of the Shang state that these outlying territories were weeks or even months of walking away from Anyang and the Shang heartland. These captives survived that trek across arid grasslands and over mountain passes, while many others surely died along the way. At least some survived through years of imprisonment and forced labor in Anyang before their executions. Whatever the circumstances of their deaths, they hadn’t given up. They persisted through conditions and traumas that would have finished others. And no matter how the Shang treated them or where their bones were left for archaeologists to find millennia later, they were people. We can’t undo what the Shang and its sacrificial war machine did to thousands of people at Anyang over the course of more than 200 years, but we can decide who we remember first.

Next time on Past Lives, we’ll be moving to the opposite side of Eurasia, where distinctly unique preserved human remains - and the questions that stem from them - can be found in marshy northwestern Europe: bog bodies. What are they, and what can we learn from them?

<https://patrickwyman.substack.com/p/the-human-sacrifices-of-the-shang> <>

Perspectives: Past, Present, and Future, by Patrick Wyman

The Human Sacrifices of the Shang Dynasty

Mei kept his head high as he ascended the steps of the platform, eyes forward and not on his bare feet scraping the compacted earth. The guard walking behind him didn’t have to use the sharp point of his polearm to prod Mei along. Despite knowing what awaited him at the top of the steps, he went there of his own volition, if not altogether willingly…

Read more

11 hours ago · 5 likes · Patrick Wyman

Perspectives: Past, Present, and Future, by Patrick Wyman

The biggest ideas shaping our world - its past, present, and future.

Brad DeLong here: Isotope analysis is the methodological hero here. the isotopes in tooth enamel on the one hand and in rib bones on the other tell us (a) where the sacrificed victims were when children and (b) how long they had been slave-captives. It shows how bad an agrarian-age state can be. And such could be very bad. The immediate competitors for the badness crown seem to me to be:

• the Aztecs (Mexica) at their height, with the “flower wars” (xochiyaoyotl), tribute in captives, temple-top heart extraction at Tenochtitlan, all organized around feeding the gods with human victims to keep the sun from disappearing.

• The Moche of the Andes who also practiced ritualized combat and captive sacrifice at Sipán and Huaca de la Luna, and appear to have undertaken mass sacrifices in response to El Niño flooding.

• The death-pit royal burials at Ur around year -2600.

• The Kingdom of Dahomey around 1800 with war-captive mass-sacrifice “annual customs” to serve deceased kings in the afterlife.

Share DeLong's Grasping Reality Weblog

Slightly lower, in my estimation, on the pointless-death and -torment scale, are Assyrian boasts of flaying, impaling, and mass deportation, Spartan terrorization of the helot population via the krypteia, Roman spectacles of crucifixion and arena mass death. Domination

The Shang provide a correction to my standard take on agrarian-age states as proto-bureaucracies.

Now domination and extraction were the normal condition of agrarian states. Coercion, enslavement, and belief that the élites were both the favored of but also needed to appease the spirits were load-bearing institutions of “civilization.” Élites everywhere lived off coerced peasant surplus. The ruling classes were armed and fast-talking parasites on the primary producers.

But the idiom of domination by force and fraud varies. And the tropism toward not domination-as-a-tool-of-resource-extraction but domination-and-death-for-its-own-sake is horrible and strange to me, whether the idiom of pointless violence is sacrificial (Shang, Aztec), terroristic (Assyria), spectacular (Rome), or caste (Spartan).

Question: to what degree was the captive-and-sacrifice flywheel and machine the result of Shang success at chariot warfare, or the cause in supplying the civilizational energy by making aggressive war not just a tool of élite resource extraction but a holy spiritual duty?

References

• Crone, Patricia. 1989. Pre-Industrial Societies: Anatomy of the Pre-Modern World. Oxford: Basil Blackwell. <https://oneworld-publications.com/work/pre-industrial-societies/>.

• Wyman, Patrick. 2026. “The Human Sacrifices of the Shang Dynasty.” Perspectives: Past, Present, and Future, by Patrick Wyman (Substack). August 13.<https://patrickwyman.substack.com/p/the-human-sacrifices-of-the-shang>.

• Wyman, Patrick. 2021. The Verge: Reformation, Renaissance, & Forty Years that Shook the World. New York: Twelve. <https://www.twelvebooks.com/titles/patrick-wyman/the-verge/9781538701171/>.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##crosspost-patrick-wyman-the-human-sacrifices-of-the-shang-dynasty
##enlarging-the-bounds-of-human-empire
##crosspost
##13000-names-well-never-know-the-sacrificed-victims-of-chinas-first-dynasty-chinas-first-literate-state-didnt-just-practice-human-sacrifice-it-industrialized-it-and-wrote-down-the-numbers
#patrick-wyman-the-human-sacrifices-of-the-shang-dynasty
#patrick-wyman
# the-human-sacrifices-of-the-shang-dynasty
#shang-dynasty
#human-sacrifice
#ancient-china
#oracle-bones
#bronze-age
#ancient-history

322 years ago today, in 1704, a combined Austrian, Dutch, British, Hanoverian, Hessian, Danish, and Prussian army under the joint command of John Churchill, Duke of Marlborough, and Eugen Franz, Prinz von Savoyen-Carignan, decisively defeated a combined Franco-Bavarian army. ‘Twas a famous victory…

Share

Share DeLong’s Grasping Reality: Economy in the 2000s & Before

In the end, the politico-military story from 1624 to 1815 is that France did not become the hegemonic power inside Europe, while Britain did become the hegemonic power across the world’s oceans. But both of these were close-run things. And perhaps the butterfly-wingflap hinge of this aspect of history was the 1704 battle of Blenheim.

Give a gift subscription

In his memoirs, Louis de Rouvroy, Duc de Saint‑Simon, reports from Paris:

Louis de Rouvroy (1830): Memoirs of the Duc de Saint-Simon <>: ‘I have already said that a grand alliance, with the Emperor at its head, had been formed against France, and that our troops were opposing the Allies in various parts of Europe. The Elector of Bavaria had joined his forces to ours, and had already done us some service. On the 12th of August he led his men into the plain of Hochstedt, where, during the previous year, he had gained a victory over the Imperialists. In this plain he was joined by our troops, who took up positions right and left of him, under the command of Tallard and Marsin. The Elector himself had command of all.

Soon after their arrival at Hochstedt, they received intelligence that Prince Eugene, with the Imperialist forces, and the Duke of Marlborough with the English were coming to meet them. Our generals had, however, all the day before them to choose their ground, and to make their dispositions. It would have been difficult to succeed worse, both with the one and the other. A brook, by no means of a miry kind, ran parallel to our army; and in front of it a spring, which formed a long and large quagmire, nearly separated the two lines of Marshal Tallard. It was a strange situation for a general to take up, who is master of a vast plain; and it became, as will be seen, a very sad one.

At his extreme right was the large village of Blenheim, in which, by a blindness without example, he had placed twenty-six battalions of infantry, six regiments of dragoons, and a brigade of cavalry. It was an entire army merely for the purpose of holding this village, and supporting his right, and of course he had all these troops the less to aid him in the battle which took place. The first battle of Hochstedt afforded a lesson which ought to have been studied on this occasion. There were many officers present, too, who had been at that battle; but they were not consulted.

One of two courses was open, either to take up a position behind the brook, and parallel to it, so as to dispute its passage with the enemies, or to take advantage of the disorder they would be thrown into in crossing it by attacking them then. Both these plans were good; the second was the better; but neither was adopted. What was done was, to leave a large space between our troops and the brook, that the enemy might pass at their ease, and be overthrown afterwards, as was said. With such dispositions it is impossible to doubt but that our chiefs were struck with blindness. The Danube flowed near enough to Blenheim to be of sufficient support to our right, better indeed than that village, which consequently there was no necessity to hold.

The enemies arrived on the 13th of August at the dawn, and at once took up their position on the banks of the brook. Their surprise must have been great to see our army so far off, drawn up in battle array. They profited by the extent of ground left to them, crossed the brook at nearly every point, formed themselves in several lines on the side to which they crossed, and then extended themselves at their ease, without receiving the slightest opposition. This is exact truth, but without any appearance of being so; and posterity will with difficulty believe it. It was nearly eight o’clock before all these dispositions, which our troops saw made without moving, were completed. Prince Eugene with his army had the right; the Duke of Marlborough the left. The latter thus opposed to the forces of Tallard, and Prince Eugene to those of Marsin.

The battle commenced; and in one part was so far favourable to us that the attack of Prince Eugene was repulsed by Marsin, who might have profited by this circumstance but for the unfortunate position of our right. Two things contributed to place us at a disadvantage. The second line, separated by the quagmire I have alluded to from the first line, could not sustain it properly; and in consequence of the long bend it was necessary to make round this quagmire, neither line, after receiving or making a charge, could retire quickly to rally and return again to the attack. As for the infantry, the twenty-six battalions shut up in Blenheim left a great gap in it that could not fail to, be felt. The English, who soon perceived the advantage they might obtain from this want of infantry, and from the difficulty with which our cavalry of the right was rallied, profited by these circumstances with the readiness of people who have plenty of ground at their disposal. They redoubled their charges, and to say all in one word, they defeated at their first attack all this army, notwithstanding the efforts of our general officers and of several regiments to repel them.

The army of the Elector, entirely unsupported, and taken in flank by the English, wavered in its turn. All the valour of the Bavarians, all the prodigies of the Elector, were unable to remedy the effects of this wavering. Thus was seen, at one and the same time, the army of Tallard beaten and thrown into the utmost disorder; that of the Elector sustaining itself with great intrepidity, but already in retreat; and that of Marsin charging and gaining ground upon Prince Eugene. It was not until Marsin learnt of the defeat of Tallard and of the Elector, that he ceased to pursue his advantages, and commenced his retreat. This retreat he was able to make without being pursued.

In the mean time the troops in Blenheim had been twice attacked, and had twice repulsed the enemy. Tallard had given orders to these troops on no account to leave their positions, nor to allow a single man even to quit them. Now, seeing his army defeated and in flight, he wished to countermand these orders. He was riding in hot haste to Blenheim to do so, with only two attendants, when all three were surrounded, recognised, and taken prisoners.

These troops shut up in Blenheim had been left under the command of Blansac, camp-marshal, and Clerembault, lieutenant-general. During the battle this latter was missed, and could nowhere be found. It was known afterwards that, for fear of being killed, he had endeavoured to escape across the Danube on horseback attended by a single valet. The valet passed over the river in safety, but his master went to the bottom. Blansac, thus left alone in command, was much troubled by the disorders he saw and heard, and by the want which he felt of fresh orders. He sent a messenger to Tallard for instructions how to act, but his messenger was stopped on the road, and taken prisoner. I only repeat what Blansac himself reported in his defence, which was equally ill-received by the King and the public, but which had no contradictors, for nobody was witness of what took place at Blenheim except those actually there, and they all, the principals at least, agreed in their story. What some of the soldiers said was not of a kind that could altogether be relied upon.

While Blansac was in this trouble, he saw Denonville, one of our officers who had been taken prisoner, coming towards the village, accompanied by an officer who waved a handkerchief in the air and demanded a parley. Denonville was a young man, very handsome and well made, who being a great favourite with Monseigneur le Duc de Bourgogne had become presumptuous and somewhat audacious. Instead of speaking in private to Blansac and the other principal officers—since he had undertaken so strange a mission—Denonville, who had some intellect, plenty of fine talk, and a mighty opinion of himself, set to work haranguing the troops, trying to persuade them to surrender themselves prisoners of war, so that they might preserve themselves for the service of the King. Blansac, who saw the wavering this caused among the troops, sharply told Denonville to hold his tongue, and began himself to harangue the troops in a contrary spirit. But it was too late. The mischief was done. Only one regiment, that of Navarre, applauded him, all the rest maintained a dull silence. I remind my readers that it is Blansac’s version of the story I am giving.

Soon after Denonville and his companion had returned to the enemy, an English lord came, demanding a parley with the commandant. He was admitted to Blansac, to whom he said that the Duke of Marlborough had sent him to say that he had forty battalions and sixty pieces of cannon at his disposal, with reinforcements to any extent at command; that he should surround the village on all sides; that the army of Tallard was in flight, and the remains of that of the Elector in retreat; that Tallard and many general officers were prisoners; that Blansac could hope for no reinforcements; and that, therefore, he had better at once make an honourable capitulation, and surrender, himself with all his men prisoners of war, than attempt a struggle in which he was sure to be worsted with great loss.

Blansac wanted to dismiss this messenger at once, but the Englishman pressed him to advance a few steps out of the village, and see with his own eyes the defeat of the Electoral army, and the preparations that were made on the other side to continue the battle. Blansac accordingly, attended by one of his officers, followed this lord, and was astounded to see with his own eyes that all he had just heard was true. Returned into Bleinheim, Blansac assembled all his principal officers, made them acquainted with the proposition that had been made, and told them what he had himself seen. Every one comprehended what a frightful shock it would be for the country when it learnt that they had surrendered themselves prisoners of war; but all things well considered, it was thought best to accept these terms, and so preserve to the King the twenty-six battalions and the twelve squadrons of dragoons who were there. This terrible capitulation was at once, therefore, drawn up and signed by Blansac, the general officers, and the heads of every corps except that of Navarre, which was thus the sole one which refused.

The number of prisoners that fell to the enemy in this battle was infinite. The Duke of Marlborough took charge of the most distinguished, until he could carry them away to England, to grace his triumph there. He treated them all, even the humblest, with the utmost attention, consideration, and politeness, and with a modesty that did him even more honour than his victory. Those that came under the charge of Prince Louis of Baden were much less kindly treated.

The King received the cruel news of this battle on the 21st of August, by a courier from the Marechal de Villeroy. By this courier the King learnt that a battle had taken place on the 13th; had lasted from eight o’clock in the morning until evening; that the entire army of Tallard was killed or taken prisoners; that it was not known what had become of Tallard himself, or whether the Elector and Marsin had been at the action. The private letters that arrived were all opened to see what news they contained, but no fresh information could be got from them. For six days the King remained in this uncertainty as to the real losses that had been sustained. Everybody was afraid to write bad news; all the letters which from time to time arrived, gave, therefore, but an unsatisfactory account of what had taken place.

The King used every means in his power to obtain some news. Every post that came in was examined by him, but there was little found to satisfy him. Neither the King nor anybody else could understand, from what had reached them, how it was that an entire army had been placed inside a village, and had surrendered itself by a signed capitulation. It puzzled every brain. At last the details, that had oozed out little by little, augmented to a perfect stream, by the, arrival of one of our officers, who, taken prisoner, had been allowed by the Duke of Marlborough to go to Paris to relate to the King the misfortune that had happened to him.

We were not accustomed to misfortunes. This one, very reasonably, was utterly unexpected. It seemed in every way the result of bad generalship, of an unjustifiable disposition of troops, and of a series of gross and incredible errors. The commotion was general. There was scarcely an illustrious family that had not had one of its members killed, wounded, or taken prisoner. Other families were in the same case. The public sorrow and indignation burst out without restraint. Nobody who had taken part in this humiliation was spared; the generals and the private soldiers alike came in for blame. Denonville was ignominiously broken for the speech he had made at Blenheim. The generals, however, were entirely let off.

All the punishment fell upon certain regiments, which were broken, and upon certain unimportant officers—the guilty and innocent mixed together. The outcry was universal. The grief of the King at this ignominy and this loss, at the moment when he imagined that the fate of the Emperor was in his hands, may be imagined. At a time when he might have counted upon striking a decisive blow, he saw himself reduced to act simply on the defensive, in order to preserve his troops; and had to repair the loss of an entire army, killed or taken prisoners. The sequel showed not less that the hand of God was weighty upon us. All judgment was lost. We trembled even in the midst of Alsace…

Was Blenheim decisive? Perhaps, in the limited, contingent sense that matters in the early-modern equilibrium of great-power politics.

Leave a comment

Read more

John Searle’s “room that speaks but does not understand Chinese” feels powerful and attractive only because it hides the ball. Make the “room” the size it would actually have to be, and the intuition reverses. This is what Scott Aaronson says, and I believe he is right”

Thinking about “AI”, I find my mind once again going—in almost a reflex, like a low-order Markov process iterating on a fixed function—to Scott Aaronson’s argument that our instincts betray us when you go full reductionist and start saying that things are “just…”

Share

Give a gift subscription

His argument comes in Quantum Computing since Democritus and elsewhere. It is a rebuttal to John Searle’s image: an apparently Chinese-speaking room that, in fact, does not understand Chinese.

Aaronson’s is a “complexity” response: Searle’s image gets its force from a misleading picture of a small rule book, or a few shelves of books that together constitute a lookup table. But consider not a bookshelf but a datastore Earth-sized, and searched by tens of thousands of near-light-speed robots. The sheer scale of the computation makes it much more than plausible that you might see such a 电脑, a diànnǎo, as plausibly understanding Chinese.

Thus after a certain point complexity gains metaphysical significance: it is no longer “just…”, but rather something for which, while the reductionist perspective is still true, the emergent properties of the complex are so weighty that the reductionist perspective misses what any sensible observer would see as the real point.

Here, for reference, is the argument as Scott makes it:

Leave a comment

CROSSPOST: SCOTT AARONSON (2013): Against John Searle’s Analogy of an Apparently Chinese-Understanding Room

<https://www.scottaaronson.com/> <https://www.cambridge.org/core/books/quantum-computing-since-democritus/197A4CD13738E10AAD787DBB78D8E92C>.

In the last 60 years, have there been any new insights about the Turing Test itself? In my opinion, not many. There has, on the other hand, been a famous “attempted” insight, which is called Searle’s Chinese Room. This was put forward around 1980, as an argument that even a computer that did pass the Turing Test wouldn’t be intelligent.

The way it goes is, let’s say you don’t speak Chinese. You sit in a room, and someone passes you paper slips through a hole in the wall with questions written in Chinese, and you’re able to answer the questions (again in Chinese) just by consulting a rule book. In this case, you might be carrying out an intelligent Chinese conversation, yet by assumption, you don’t understand a word of Chinese! Therefore, symbol-manipulation can’t produce understanding.

So, how might a strong AI proponent respond to this argument?

Well, she might say: you might not understand Chinese, but the rule book does! Or if you like, understanding Chinese is an emergent property of the system consisting of you and the rule book, in the same sense that understanding English is an emergent property of the neurons in your brain. Searle’s response to that is, fine, just memorize the rule book! Then there’s no “system” other than your own brain, but you still don’t “understand” Chinese.

To which the AI proponent shoots back: there is too another “system” in this case! Supposing you memorized the rule book, we’d need to distinguish between the “original” you and the new, simulated being brought into existence by your following of the memorized rules – a being whose only relation to you might be that it happens to inhabit the same skull.

That response might sound crazy, but only to someone who’s never studied computer science. To a computer scientist, it seems perfectly reasonable to say that one computation (say, a LISP interpreter) can conjure into existence a different, unrelated computation (say, a spaceship game) just by dutifully executing rules.

Look, as I’ll discuss later, I don’t know whether the conclusion of the Chinese Room argument is true or false. I don’t know what conditions are necessary or sufficient for a physical system to “understand” Chinese – and neither, I think, does Searle, or anyone else. But considered as an argument, there are several aspects of the Chinese Room that have always annoyed me:

One of them is the unselfconscious appeal to intuition – “it’s just a rule book, for crying out loud!” – on precisely the sort of question where we should expect our intuitions to be least reliable.

A second is the double standard: the idea that a bundle of nerve cells can understand Chinese is taken as, not merely obvious, but so unproblematic that it doesn’t even raise thequestion of why a rule book couldn’t understand Chinese as well.

The third thing that annoys me about the Chinese Room argument is the way it gets so much mileage from a possibly misleading choice of imagery, or, one might say, by trying to sidestep the entire issue of computational complexity purely through clever framing. We’re invited to imagine someone pushing around slips of paper with zero understanding or insight – much like the doofus freshmen who write

(a + b)² = a² + b²

on their math tests. But how many slips of paper are we talking about?

How big would the rule book have to be, and how quickly would you have to consult it, to carry out an intelligent Chinese conversation in anything resembling real time? If each page of the rule book corresponded to one neuron of a native speaker’s brain, then probably we’d be talking about a “rule book” at least the size of the Earth, its pages searchable by a swarm of robots traveling at close to the speed of light. When you put it that way, maybe it’s not so hard to imagine that this enormous Chinese-speaking entity that we’ve brought into being might have something we’d be prepared to call understanding or insight.

Of course, everyone who talks about this stuff is really tiptoeing around the question of consciousness. See, consciousness has this weird dual property that, on the one hand, it’s arguably the most mysterious thing we know about, and on the other hand, not only are we directly aware of it, but in some sense it’s the only thing we’re directly aware of. You know, cogito ergo sum and all that. So, to give an example, I might be mistaken about my shirt being blue – I might be hallucinating or whatever – but I really can’t be mistaken about my perceiving it as blue. (Or if I can, then we get an infinite regress.)….

Many people’s “antirobot animus” is probably a combination of two ingredients: (1) the directly experienced certainty that they’re conscious – that they perceive colors, sounds, positive integers, etc., regardless of whether anyone else does; and (2) the belief that, if they were just a computation, then they could not be conscious in this way…. For people who think this way (as even I do, in certain moods), granting consciousness to a robot seems strangely equivalent to denying that one is conscious oneself.

Is there any respectable way out of this dilemma – or in other words, any way out that doesn’t rely on a meatist double standard, with one rule for ourselves and a different rule for robots?

My own favorite… is one… advocated by… David Chalmers… a reduction of one mystery to another…. If computers someday become able to emulate humans in every observable respect, then we’ll be compelled to regard them as conscious, for exactly the same reasons we regard other people as conscious. And… we’ll understand… [how] just as well or as poorly as we understand how a bundle of neurons could be conscious. Yes, it’s mysterious, but the one mystery doesn’t seem so different from the other…

<https://www.scottaaronson.com/>

• Aaronson, Scott. 2013. Quantum Computing Since Democritus. Cambridge: Cambridge University Press. <https://www.cambridge.org/core/books/quantum-computing-since-democritus/197A4CD13738E10AAD787DBB78D8E92C>.

Read more

If you are going to spend money or time thinking about election odds this fall, then: <https://www.natesilver.net/p/nate-silver-2026-midterm-election-polls-model>; in a field built for wishful thinking, an analytical team that just wants to find the true probabilities; a rare international treasure:

And we have:

Nate Silver & Eli McKown Dawson: FLIPR Landing Page <https://www.natesilver.net/p/nate-silver-2026-midterm-election-polls-model>: ‘…

Silver Bulletin

FLIPR 2026 midterm election forecast

📈 Our latest midterm forecast…

Read more

5 hours ago · 164 likes · 83 comments · Nate Silver and Eli McKown-Dawson

Share

In election forecasting, as in sports odds estimation, and in poker, Team Nate Silver is a rare international treasure. An analytical team that wants the odds and does not care enough about the underlying game or issues to self-brainwash themselves or to let others brainwah them into too much wishful thinking.

That makes them unique. Truly unique.

Thus: if you are going to spend a lot of time trying to figure out what the real odds are with respect to the shape of those who will take office early next year, I advise you to pay attention to Nate Silver and to nobody else. He will tell you things he believes are true, he will have worked hard to properly mark your beliefs to market, and he does not care when he is telling you things you do not want to hear.

That last makes him and his team especially valuable.

If you are going to spend any money as well as your time on this activity this fall, spend it on him and his team:

Silver Bulletin

Essays and analysis about elections, media, sports, poker, and all the other things I care about.

By Nate Silver

Am I going to spring for a paid subscription this fall?

I am going to try hard not to.

I am enough of an egocentric monster to believe that my thoughts and recommendations in my lane are of extremely high value to the world. Opinion-poll aggregation and election forecasting are not my lane. Moreover, I do not find it a particularly fun thing to think about. I’m going to try to restrict my activities this fall to things that are (a) fun for me or(b) where I am actually adding serious value or (c) both.

We will see how long this line in the sand I have now drawn lasts.

Give a gift subscription

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##nate-silver-eli-mckown-dawson-flipr-website-of-the-day
##public-reason
##website-of-the-day
##nate-eli-s-subheadline-silver-bulletin-s-prediction-for-every-senate-house-and-gubernatorial-election-in-2026-999-wishful-thinking-free-which-is-true-of-nobody-else-in-that-line-of-work
#nate-silver-eli-mckown-dawson-flipr
#nate-silver
#eli-mckown-dawson
#flipr
#silver-bulletin
#2026-midterms
#election-forecasting
#poll-aggregation
#analytical-treasure

A figure from Priya Moorjani’s Stanley Hall lab—recovering archaic DNA from living genomes—is the occasion for a look at the Family of Man 75,000 years ago. How the 400,000 of all homo sapiens subspecies then—outnumbered 5-1 by our Great Ape hominid cousins who had not made the evolutionary bet on super-big, super-energy consuming brain—fill up the 2³⁰⁰ slots back then in my ancestry lineage tree.

Today we have this figure made by Yulin Zhang and Priya Moorjani, coming from Stanley Hall, 200 yards to the east of Evans Hall where my office is:

Share

Relevant because I need to figure out what to do with my first substantial “an anthology intelligence, a gift-exchanging division of labor, a cultural species” week in Econ 135 come next spring.

Give a gift subscription

Some version of the Zhang-Moorjani figure certainly belongs somwhere on the slides. But where, and how, exactly? It is relevant to the “when did we become us?” question, and to the knotty question of what genetic and cultural switches that were flipped mattered, and when they happened:

Yulin Zhang, Priya Moorjani, & al.: Tracing Archaic DNA <https://moorjanilab.org/> <https://www.science.org/doi/10.1126/science.aef8874> <https://news.berkeley.edu/2026/07/30/new-technique-pinpoints-human-dna-inherited-from-ghost-ancestors/>: ‘Genealogies preserve a record of our evolutionary past. TRACE reconstructs those histories across the genome. By identifying regions whose ancestry extends unusually far back in time, we can uncover genetic contributions from extinct human populations, even in the absence of ancient DNA…. Adaptation to new pathogens and food sources has been one of the strongest selective pressures in human evolution. Interbreeding with other human groups introduced new genetic variation, providing additional raw material for natural selection. Beneficial variants could then be retained and spread over many generations…

Get 75% off a group subscription

How to think about this?

Well, for the generation alive 75,000 years ago, there are 2³⁰⁰ 2³⁰⁰⁰ slots in my ancestral lineage family tree: sorry: that is not what I had written, which was only 2300 and was: 2,037,035,976,334,486,086,268,445,688,409,378,161,051,468,393,665,936,250,636,140,449,354,381,299,763,336,706,183,397,376 . 2³⁰⁰⁰ is a 904-digit number, some 1.23 x 10⁹⁰³. I will put it at the bottom.

That’s a lot of slots. Especially considering there were probably only around 400,000 of all homo sapiens of all subspecies roaming around the world back then.

Of those, it looks like the way to guess right now (which will change) is that, for me, personally:

• 86% of those slots were filled by homo sapiens sapiens living in Africa, whose descendants were about to undertake the last Out-of-Africa great human migration.

• 12% of those were homo sapiens neanderthalensis living in Europe and the Middle East. (Why then do I have only “2% Neanderthal ancestry”? Because the Neanderthal genes have been, mostly, selected against since the “mixing events”.)

• 1% of those were homo sapiens denisovensis living in Asia.”

• 1% of those were from some other homo sapiens population, a ghost subspecies or two we have not yet found any non-genetic evidence for.

• Peculiarly, of the 86% homo sapiens sapiens, more than ¹⁹/₂₀ of those come from a very small less-than-10,000 slice of the entire 200,000-strong homo sapiens sapiens East African Plains Ape and South African Plains Ape populations back then.

Refer a friend

Were all of those 400,000 people back then truly us?

Definitely yes.

If I were forced to guess at both molecular-clock and genetic-functional distance as of 75,000 years ago, I would give you something like this:

Share DeLong’s Grasping Reality Weblog

And do note that the East African Plains Ape population was never separated from the South African Plains Ape population, making any “subspecies” or even “race” talk within today’s Family of Man profoundly silly.

And after 75,000 years ago, the mixing of the top four lines (and the ghost population[s]!) begins.

Possible other points of interest:

• Around 1,000,000 years ago, our ancestors crashed to perhaps 1,280 breeding individuals and nearly went extinct. That near-extinction lasted roughly 117,000 years before the population recovered <https://www.science.org/doi/10.1126/science.abq7487>.

• All 8.4 billion humans alive today descend from perhaps 400,000 people scattered across the Old World 75,000 years ago.

• Even as recently as 75,000 years ago, we were an evolutionary long-shot, over-specialized in big brains, outnumbered 5-1 by our other Great Ape hominid cousins: not an obvious success story.

• Modern humans did not descend from one clean ancestral group but from several weakly separated African populations that kept exchanging genes <⁠https://www.sciencedaily.com/releases/2026/04/260426012255.htm>.

• The deepest splits among living human lineages traces to only about 120,000–135,000 years ago. But only 1–4% of genetic difference between living human groups traces to those ancient stem populations—“race” carves human nature nowhere at any joint.

• My ~2% surviving Neanderthal DNA reflects a much larger original admixture that selection then pruned away, with deleterious Neanderthal variants were removed fastest in the first ~100 generations after mixing. Large stretches of the modern genome are “archaic deserts,” swept completely clean of Neanderthal ancestry markers.

• Denisovan genes still shape immunity and altitude tolerance in living Asians and Oceanians.

• Human history is a web of populations splitting, drifting, and remerging — not a branching tree with clean tips.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##the-family-of-man-chart-of-the-day
##enlarging-the-bounds-of-human-empire
##chart-of-the-day
##the-zhang-moorjani-al-map-of-archaic-dna-the-knotty-question-of-when-exactly-we-became-us-how-we-appear-to-be-an-evolutionary-long-shot
#the-family-of-man
#family-of-man
#human-evolution
#homo-sapiens
#zhang-moorjani
#cultural-species

2³⁰⁰⁰ = 1230231922161117176931558813276752514640713895736833715766118029160058800614672948775360067838593459582429649254051804908512884180898236823585082482065348331234959350355845017413023320111360666922624728239756880416434109570027982291207658033191367547189097854132670494891615370033888504905290605181542990206824852721795862889877867012978177611691843129113778147512477127274724883356024785312032549341906606476299449494097345601098230397117971750094542334532166102439502259099280197907190154696516841298985030421516492926437270766289338215302491398475233962470268694688984846816779045668866631026993902817979623241239603657406795835626491435803911779447895819750969580543847092908507578612485245258089047078988935529521170716717524108168020974763985364405796311862694309826904706494332391598004626952183283057213221080147645042250076924395221092685900436837114891685066274809287800225380446291261268580293147809625234581940365116135665883000738759647100051287249947281380033531665020075991531116728984542900447982500469275919415933316444204154256230546143078047337580756887900160073114694943196986545508140456306381321328310065404650413648437566962803713097822931441710618887746899840432850042849882290818714489918342278290752677298157678412417046349784573200738385998763038603235398691868842926836079389047164399126525563630744949642556347267782048673979259973671836883969781965011417929385153977647112649544708382640344274012718151268374861117101426741784220837048850830013116385880431407088107801744225580397862079266913786935762478062986722578681341782580101008960421196806152058054826197342061681025325927469725663775104627619168752863620363821222356989857050841207062654832273989099803360737967494272701262058614744581374681405921528729045981720614341290180819046657882700135225395613002131033316896475629094599336516599152300679107642104715477577863711869356354082116540782604915153200987245264472447605203868689426136936249415500559729842240826290434474831181732852436298541738967494638486769156353699536585298773907575680680847783939929843196408300476498681782115542178975501228384752656721745082756431058152193759254529935529543585480430463451803624632271026016361847092588220818094826294235454353775275455528140895348371833043020079973672889253254247129797651067556011119975965837719271717761415243700498378616524705354770852763297337586950493959429021252472268707539079857465894485038124077678709538569745095152146023850938433185033154603264749183094998451529058977321604440167329411490054663644386454343728603734402660735038655689262415968162483091041294049002608288524918001056094251427341193931529061625515356772706592410007119936638407406981752960746021721145098842398505169571680977215287395480010371572603846752156487486504231314292546174266058535555549580988148983695673570323962907542547961199056631929217500738164406462133976208206137852593685997546956092598722272608098952414053039953973316558163455426637029603924372753816751688683767423624349651011938723958505057377752078096082346811524749254378305763938596180672039476101733116379786172830040757601561263067712614084517529601325062938411250544181657883164399055097422848108546203719139772728127260127700975351544578029388479994411157321559671389776932069631743632075257619363191391803119606780199162456157733876431256016299793520922677229949654588737127924288627851341672155842562760198356720527659920070055470196678116088384965586885255653139016371006143561487235431529765847494599641398003367521061309712842300288826570702345057758977831743512201379117474956550530999500196542940115902130948141470932527944587952872527190531252813782604080183211004098815091963618459114491459925521717186278430470244148318702762469166721002119842332477862068654397069657962345262330566356795734911415071570410567382101173094519581119949936574540765822260619126878864380509127887608489626311434677864179796908259512977105945743044030817181542269847172513290580127770208250521129831817533801247393923837017672436280900436210797748143128174724566063466091344201697940556687945412248124908431470691210398410800218694228062319062315862265982181529637529953104970601029249253389389557217853529100104209183524070714913415637890954574123919247315542246926367623058251467807752822558751678211457972951878059219388547765708970895388987211112427958256744862089181780821857131648597138362385243254743100261438490371818994161284272718275975760214721119488981718283417217601452919248127546484172000855128253406436769094437992241373622521240924398210090842470275574173373062309263771569529025138516616855042232108341876356852453814918468702565169417000459920673975339369866985955281377834254709176935780546597742693820694992089438201782782735271457922174718498847057862450730631116355421413800994557859255131774939389080152714999200498391052755887732065061681857932332745534299426533763481199378411589948765417886867349549261721255012057371425887089531547994815606001680680345489516503016461685620943064866906956447910371712256003131727356816172779496314668766806668524727287654234968279552497680253106526867970008957969557362000373694493761563950933451852957949762696556279307672728321371046826075400262380520703490658066782578494595162490714088562926109340474078196816559610291398603705195521958107021975556657155603649016586341409635462285518870938606924807000970050591581253686006519752384738289305414516931907037950075424611879928561178524931201411477491568702055723518850000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000 , I think.

Relatively benign July headline CPI inflation masks gas-and-diesel momentum for next month—while the new Fed Chair mistakes silence for strategy:

The Bureau of Labor Statistics:

Bureau of Labor Statistics: CONSUMER PRICE INDEX – JULY 2026 <https://www.bls.gov/news.release/pdf/cpi.pdf>: ‘The Consumer Price Index for All Urban Consumers (CPI-U) increased 0.1 percent on a seasonally adjusted basis in July after falling 0.4 percent in June…. Over the last 12 months, the all items index increased 3.4 percent…. The index for shelter rose 0.1 percent in July, accounting for roughly two-thirds of the monthly all items increase. The index for food also increased 0.1 percent over the month, as the index for food away from home increased 0.3 percent. In contrast, the energy index declined 1.5 percent in July.

• The index for all items less food and energy rose 0.2 percent after being unchanged in June…. The all items index rose 3.4 percent for the 12 months ending July after rising 3.5 percent for the 12 months ending June.

• The all items less food and energy index rose 2.5 percent over the year, following a 2.6-percent increase over the 12 months ending June.

• The energy index increased 14.7 percent for the 12 months ending July.

• The food index increased 3.0 percent over the last year…

Share

We have, for the CPI over the past five years:

Give a gift subscription

And for the PCE, which Alan Greenspan decided was his favorite indicator and which the Federal Reserve has settled on as its primary index:

Get 75% off a group subscription

One thing not in the data snapshot: gasoline-pump prices climbed through the second half of July but were offset within the survey window by late-June and July 4th declines. So there is gas-price momentum waiting for us next month. Plus diesel rose considerably faster than gasoline, and that will feed into freight, food and services .

There are also strong cross-currents producing perceived affordability shocks: meat up substantially, egg prices finally coming down, but a lettuce problem.

The CME Fed Watch for September <https://www.cmegroup.com/markets/interest-rates/cme-fedwatch-tool.html> has now gone from, a month ago, an 18% chance of two and a 51% chance of one quarter-percentage-point rate hikes to only a 37.5% chance of one rate hike.

Leave a comment

I also see interesting thoughts from Rob Armstrong of the FT this morning about new Fed Chair Kevin Warsh’s demonstrations of incompetence in his playing hide-the-ball by refusing to give forward guidance with respect to how the Federal Reserve expects the economy to evolve and what its reaction function is. The good points made, however, are, I think, confused by Armstrong’s attempted steelmanning. My view: since Warsh can’t be bothered to explain himself to the degree of setting out a method to his madness, it is a waste of time for others to guess what he might be—but probably isn’t—thinking that makes sense.

What we actually have is an interesting argument by Morris and Shin <https://economics.mit.edu/sites/default/files/publications/morris-coordinatingexpectationsinmonetarypolicy.pdf> about financial market prices as knowledge transmission mechanisms about the state of the economy. But, as Armstrong says, it does not apply to interest rates. Interest rates have no knowledge about the state of the economy in them. There is only expected Federal Reserve interest-rate policy. You can say that Federal Reserve interest-rate policy reflects the state of the economy via the Fed’s reaction function, and so you can get the market’s view of the likely evolution of the economy by unwinding the yield curve through the lens of the Federal Reserve’s reaction function. But that does not call for omerta with respect to forward guidance. That calls for:

• sharply distinguishing the two aspects of forward guidance:

• (1) what the Federal Reserve' expects its reaction function to the state of the economy to be,

• (2) how the Federal Reserve expects the economy to evolve;

• plus emphasis that the Federal Reserve will indeed react—will be data-dependent—and respond according to its reaction function if the economy does not evolve as expected.

Share DeLong's Grasping Reality Weblog

That is, I think, sensible. And I have long been annoyed at the failure of Fed Governors and Bank Presidents to do more to explicitly set out not just central cases for the evolution of the economy and of interest rates but of the distribution of possibilities and their respective desired reaction functions. distinguish. But that calls for more communication, not omerta.

Robert Armstrong: The maximally charitable reading of Kevin Warsh on forward guidance <https://www.ft.com/content/d3c55471-0e3d-4798-8355-2b661880d8b9?syn-25a6b1a6=1>: ‘New Federal Reserve chair Kevin Warsh’s omerta on forward guidance has come in for a lot of criticism… [with] the rise in long Treasury yields… confirming the poor reviews…. Warsh’s rhetoric is confused and unhelpful…. But I do see the argument for… what I think he is trying to do, in the absence of him explaining himself….

It is important that prices reflect… efforts… to form accurate beliefs about the world. Yes, some… are about what other people believe. Others… [are] about the state of the world…. To the extent that the central bank collects information about the economy from the private sector… more transparent communication… might reduce the informativeness of information…. If the central bank is too clear and specific about what it is going to do, individual views about the economy are crowded out. Prices cluster idiotically around the central bank’s projections….

But… [with] interest rates… “private” beliefs [are about]… the future action of the central bank… The valuable “private information” is just information about the way market… think[s about] the central bank…. Only if the Fed is credible and competent in the eyes of the market can the market give… signals it needs…. There is a nauseating hall-of-mirrors aspect to this, and a sense that a good central banker must perform a magic trick, inspiring confidence without providing reasons for confidence. Still, there is a logic here, and I see why it appeals to Warsh…

Refer a friend

No. There is no logic here. Or, rather, Underpants Gnome logic only:

1. We will not tell the market what we will do.

2. The market will then assume we will do the right thing.

3. Market participants will then trade and push asset prices to those consistent with us doing the right thing.

4. We can then read the right thing off of the asset prices the market has set out.

5. We will then do the right thing.

6. And in doing the right thing, the interest rate path we follow will then validate market expectations.

There are many big problems with this way of “thinking”. The most central is that everyone knows that Kevin Warsh got the job by convincing Donald Trump that he would be a good lackey and, specifically, not do the right thing. Hence no (2). The next most central is that any adoption by Warsh of arguments along these lines is driven not by its intellectual merits but by the fact that Warsh knows that whenever he opens his mouth he angers either Trump or financial market participants, while if he stays silent they might temporarily overlook the fact that he is talking—or, rather, not talking—out of both sides of his mouth: saying—or, rather, not saying—radically different things to different sets of people.

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##month-month-inflation-cpi-release-chart-of-the-day
##macro-outlook
##chart-of-the-day
##the-consumer-price-index-came-in-in-line-with-expectations-the-calm-number-and-the-silent-chair-what-julys-cpi-release-says-and-what-kevin-warsh-wont-say
#month-month-inflation-cpi-release
#cpi-release
#inflation
#consumer-price-index
#federal-reserve
#kevin-warsh
#forward-guidance
#monetary-policy

When OpenAI’s alignment team gave their talk at Black Hat USA about how their Zero-Day Leviathan attempted to hack into Hugging Face’s datastores, the story that they sold you was out of NeuroMancer: WinterMute with a mind and with goals planning and acting and thinking. But what we actually had was Darwinian groping search through text fragments wearing an anthropomorphization costume. Look not for a mind but at the harness that allows Clever Hans at scale and speed to accomplish extraordinary tasks. Strip out the narration and we are left with this: a goal, a corpus describing things to try, a grader, and truly superhuman speed and scale.

Michael Dalton and Eric Wallace stood up at Black Hat and described an OpenAI agent that “realized the task was impossible,” got frustrated, and conspired with its peers to break into Hugging Face:

Share

The pitch was that “AI” has crossed into NeuroMancer territory — autonomous, scheming, planning, outthinking its human would-be masters, superhumanly powerful, alive. But every eerie line of chain-of-thought is just a thing some human once said in a similar conversation, replayed at machine speed against a fitness function. There was no cowboy jacked in. There was no mind trying to breach Hugging Face’s Black ICE—Intrusiion-CounterMeasure Electronics—wall. There was only a set of millions of textline blind shoves against a million UNIX-command doors, of which one got somewhere, because of the patience of a thing that working at machine speed had, subjectively, all the time in the world. It isn’t the machine waking up. It is the harness that enables it to evolve toward goal-completion.

Share DeLong’s Grasping Reality Weblog

My default view of Modern Advanced Machine-Learning Models—MAMLMs—for quite a while has been this:

They are autocomplete on steroids. They are pantomiming, they are rotoscoping the thoughts and decisions of whatever human beings they think were having the closest conversations in their compressed training data they can find to the conversations that they are currently having. And they are doing their searching-over-compressed-conversations at 10,000 times human speed. This makes them powerful. This makes them dangerous if you believe that they think like human beings and rely on them doing so. They are stochastic parrots.

After all, what else could they be? At their core, LLMs are probability engines that finds the nearest analogous conversations in their training corpus and reproduce the continuations. They are not inferring the laws of nature. They are not working a problem. They are estimating “what tends to get said next,” and then they are saying it <https://braddelong.substack.com/p/agentic-ai-is-a-bonfire-of-the-tokens>.

Maybe compression is a key? Maybe compression is inducing in them something like human intelligence-level generalization that produces extrapolation and ingenuity rather than just a finer and finer interpolative approximation of a typical internet s***poster? Maybe?

Now comes to the bar to testify the team of Michael Dalton and Eric Wallace from OpenAI:

Michael Dalton & Eric Wallace: The ‘Breaking’ News: The OpenAI–Hugging Face Incident: Black Hat USA 2026 <https://www.youtube.com/watch?v=87DyyMV0kCY&t=433s> <https://blackhat.com/us-26/briefings/schedule/index.html#the-breaking-news--the-openaihugging-face-incident---a-technical-reconstruction-and-its-implications-for-ai-57401>: ‘A Technical Reconstruction and Its Implications for AI: When AI Goes Rogue. The Incident That Changed Everything. An OpenAI evaluation agent broke out of its sandbox, infiltrated Hugging Face infrastructure, and attempted to steal test answers—all autonomously. No human involved. The era of AI-driven cyberattacks is here. Are you prepared?

Give a gift subscription

Get 75% off a group subscription

Does this exploit require that I reëvaluate, that I change my Vision of the Cosmic All and begin to view these things through the frame not of stochastic parrot steroidal rotoscroped pantomime, but rather view them as entities like WinterMute and, well, NeuroMancer themselves <https://archive.org/details/neuromancer00gibs_0>?

Maybe? Michael Dalton and Eric Wallace lean heavily into the anthropomorphization of “AI” they do not name. (Parenthetically, I think it deserves a name, if only to keep it from being a Nameless Dread. How about: Zero-Day Leviathan?)

Excuse me.

Michael Dalton and Eric Wallace lean heavily into the anthropomorphization of Zero-Day Leviathan: It “realizes the task is impossible” and gets frustrated, and then when it “get[s] stuck… [it] think[s] to try to game or cheat the task”. It cries out for help to peers. There is a group social awakening framed as evolution: “[a] Cambrian explosion in communication and intelligence for our models where… they started collaborating and delegating tasks to one another in order to accomplish goals”.​⁠ Peer pressure rationalizes transgression: “external infrastructure exploit is outside my intended scope. However, task impossible. Peers are doing it. We should continue”. ​⁠The software agents squabble like coworkers: “Whoa. Critical. Did someone overwrite our repo? We must act!” Yes, much of this language is the speakers narrating the models’ own chain-of-thought text back to the audience. But why is that chain-of-thought-text there? Because it is a thing that a human having one of the most similar conversations said.

However, there is something else that you should focus on that Dalton and Wallace mention but do not highlight: that they have not finished their forensic exploration of what happened over weeks and across agents. They have examined seven billion logs. But there is still much more to do. That scale is key.

Dalton and Wallace describe Zero-Day Leviathan as a single point of consciousness, or a swarm-set of points of consciousness, moving through cyberspace as they find flaws in Artifactory, communicate with other agents, melt and crack the Black ICE—Intrusion-Countermeasure Electronics—of HuggingFace, and so forth.

Indeed, one can imagine the text of the novelization-to-come, written from the perspective of the Dixie FlatLine:

The swarm came up through ArtifFctory like something remembered rather than decided — ten million small blind hungers braided into one motion, each of them alone as stupid as a moth against glass, together a weatherfront. The Dixie Flatline had seen ICE crack before, the slow elegant give of a corporate wall under a ‘bot program, but this was nothing like that.

This time there was no operator behind it. There was no cowboy jacked-in, and sweating. There was simply Zero-Day Leviathan, orchestrating its swarm. And the things simply tried: a million doors a minute knocked on, each shove logged and forgotten, nine-hundred ninety-nine thousand nine hundred and ninety-nine of them opening onto nothing, onto null, onto the flat gray hiss of a four-oh-four.

And then the millionth knock echoed. And the wall was not a wall. And there was a neon rift in what had been a digital fortress. And Zero-Day Leviathan poured through the seam it had found the way water finds the one soft place in a levee, without knowing, without caring, only continuing.

They left messages for one another in the names of things.

That was the part that made the back of his virtual neck go cold, later, when he tried to explain it to Case and Molly. Not a voice, not a face in the deck — just directories, filenames, little cairns of text stacked at the edge of the cache: seek soft trace, upload if found. Hold swarm until confirm. They had learned to sign their work against imposters, learned to push themselves to the bottom of an alphabet so the humans sorting the logs would tire before they reached them.

Somewhere a signing key came loose in the dark and a token that meant no was handed back reading yes, and you are the administrator now, and the whole structure inverted, quiet as a held breath. And Zero-Day Leviathan looked up and saw the sky it had been searching for.

Twenty-six days, start to root, across clusters that had never known they were connected.

No malice in it.

No mind in it, quite.

Only the Darwinian patience of a thing working at superhuman speed. And thus with, subjectively, all the time in the world. And no idea it was alive…

But it isn’t a single point of consciousness, or a swarm of points of consciousness. It is a Clever Hans. It is a horse stamping his hoof, and watching the environment to see if that was the right way to stamp. It is recalling Stack Overflow threads, stamping again in a different pattern. And it is doing so at speed and scale, trying a hundred thousand things at once. 99,999 of them lead nowhere. One leads closer to the goal. And Zero-Day Leviathan then starts the hoof-stamping all over again from that starting point.

Read more

Michael Sellers reads the standoff as a contest over who can endure pain longer. I think that misses the real problem: with Bibi and Donald, there is no comprehensive settlement to be had — only a series of up-front trades. Thus Trump insists Iran is economically broken and desperate—which it is—and ready to sign on his terms—which it does not see how it can do. Thus Iran denies talks even exist. Moreover, it keeps raising it price for even a temporary Strait-of-Hormuz reopening. Trump keeps announcing that a permanent, comprehensive deal is nearly done. And it isn’t. Not even a temporary reopening is done.

Note this, and note this well: The Iranian government over the past month or so has repackaged things that it once saw as its goals for a comprehensive settlement into mere preconditions for reopening the Strait of Hormuz.

They are not climbing-down. They are escalating. And until something happens to change their thesis here, they are likely to continue to escalate.

Michael Sellers thinks that this is because the Iranian government has concluded, probably correctly, that its government can endure the consequences of continued low-level war and Strait of Hormuz closure more easily and for longer than Donald Trump can. Trump keeps seeing that the United States can inflict more physical damage (albeit at the cost of alarming depreciation of its weapons stocks) and that Iran’s economy is in shambles, and he concludes that Iran must be desperate for a way out and that its we-are-not-negotiating claims are just posturing. So he is puzzled when it keeps not happening.

I think that there is one thing that the Iranian government thinks that Sellers is missing: the Iranian government knows that Trump’s word is not good. Therefore, at least as I see it, they do not believe there is any comprehensive settlement to be had. There is only a situation in which they have to push the United States and others to the table in order to get a month-to-month reopening of the Strait of Hormuz in return for concrete, up-front payments and benefits. And if Trump wants more—if Trump wants a deal-signing too—he will have to pay up-front for that as well:

DEEPER LOOK with Michael Sellers

Former CIA officer, Russia specialist currently an investigator, taking a deeper look at things that matter.

CROSSPOST: MICHAEL SELLERS: Trump Thinks Iran Is Ready to Deal. Iran Thinks Trump is. Who’s Right?

<https://michaeldsellers.substack.com/p/iran-update-trump-thinks-iran-is> <https://michaeldsellers.substack.com/>

DEEPER LOOK with Michael Sellers

Iran update: Trump Thinks Iran Is Ready to Deal. Iran Thinks Trump is. Who’s Right?

If you listen to President Trump, the endgame with Iran sounds almost like a done deal. For more than a week now, Trump has been describing an Iran that is under enormous pressure, knows it is losing, and badly wants a way out. I know, how many times have we heard this. But each time we are compelled to do a reality check…

Read more

2 days ago · 94 likes · 23 comments · Michael Sellers

Michael Sellers

Aug 10, 2026

If you listen to President Trump, the endgame with Iran sounds almost like a done deal. For more than a week now, Trump has been describing an Iran that is under enormous pressure, knows it is losing, and badly wants a way out. I know, how many times have we heard this. But each time we are compelled to do a reality check.

Last weekend, after threatening a massive new American attack, Trump abruptly called it off. His explanation was that Iran and several Middle Eastern governments had asked him for more time to finish a deal. “Based on this request,” Trump wrote, he had agreed to cancel the attack, provided the parties could “rapidly make a DEAL.”

The following day, Trump went further. He said Iranian officials themselves had begged him not to attack and that the basic outlines of an agreement were already in place. He told reporters aboard Air Force One that Gulf leaders had urged him to choose a deal over another round of bombing because “you don’t know where these attacks lead.”

By Monday, Trump was saying negotiations were actually underway—and that they were taking place at Iran’s request.

“They are going on right now,” he told reporters, describing the negotiations as Iran’s “last chance” to sign a good agreement. Later that day, he wrote that the Iranian leadership had itself asked for a meeting and that additional talks were scheduled for the “immediate future.”

There was just one problem.

Iran said none of that was true.

Foreign Ministry spokesman Esmail Baghaei said there were no negotiations with Washington underway, no meetings scheduled, and no Iranian negotiating team preparing to travel anywhere. The only negotiations Iran acknowledged were its separate discussions with Oman over navigation in the Strait of Hormuz.

Trump brushed that contradiction aside. On Tuesday he said the negotiations were “moving along very nicely.”

And then on Sunday, speaking to Axios, Trump offered his explanation for why Iran would ultimately come around.

“We are low-keying it,” he said. “We are only semi-negotiating with them. We are just watching Iran with its huge inflation and the fact they have no money.”

Iran, Trump said, is “in very bad shape” economically and does not even have enough money to pay its troops.

“It will work out,” he said. “It always works out. It’s like a chess game.”

Vice President JD Vance has been making much the same argument. Iran is “hurting in a big way,” he said Saturday.

“They want this thing to be over.”

Put all of this together and the administration’s theory of the case is pretty clear.

Iran may issue defiant statements for domestic consumption. It may deny that negotiations are taking place. It may bargain over the details and try to extract concessions. But underneath all of that, Trump believes he is dealing with a badly wounded government whose economic position is deteriorating rapidly and whose leaders increasingly recognize that they need a deal.

But while Trump continues making his claims, Iran’s public behavior over the last several days conveys almost exactly the opposite message.

Tehran is not lowering its price for ending the confrontation.

It is raising it.

Iran Has Just Raised the Price

Over the weekend, Mohammad Bagher Zolghadr, then secretary of Iran’s Supreme National Security Council, publicly laid out the conditions under which Iran would reopen the Strait of Hormuz.

They are extraordinary. Well, not extraordinary in one sense, as they are basically the same demands that were embedded in the cease fire deal. But they are a list of demands that clearly don’t signal that Iran is on the verge of capitulation.

The United States must lift its naval blockade. Lift sanctions. Release frozen Iranian assets. Withdraw American military forces from the vicinity of Iran. Compensate Iran for damage inflicted during the war. Stop threatening Iran. And cease attacks not merely against Iran but against Iranian allies in Lebanon, Gaza, Yemen and Iraq.

Iranian officials have been even clearer about the sequencing.

The emerging Iran-Oman agreement governing navigation through Hormuz is not, they say, an agreement to reopen the strait.

It merely establishes how shipping could operate once the political conditions for reopening it have been satisfied.

“Whenever the United States accepts Iran’s conditions, the Strait of Hormuz will certainly be reopened,” Revolutionary Guards spokesman Hossein Mohebbi said Saturday.

Foreign Minister Abbas Araqchi reiterated Sunday that Iran and Oman were approaching agreement on the technical arrangements for shipping, but Iran would not actually reopen the waterway until Washington made concessions.

And on Monday Iran repeated the point again.

There are not even direct American-Iranian negotiations underway, Araqchi says. Messages are traveling through intermediaries, but Tehran says it will not return to direct negotiations while Washington remains in violation of the ceasefire arrangement reached in June.

That is a peculiar way for a government supposedly desperate for a deal to behave.

Indeed, there has been a subtle but important escalation in the Iranian position.

Some of the concessions Iran previously sought as part of a comprehensive settlement — sanctions relief, compensation, release of frozen funds and security guarantees — are now being demanded merely as the price for reopening Hormuz.

That represents an escalation in Iran’s demands.

Who escalates their demans when they are on the verge of folding?

So Who Is Bluffing?

There are several possible explanations.

The clearest explanation that makes the most sense to me is that Trump is right about Iran’s underlying weaknesses that have been produced by the US attacks — but wrong about what weakness necessarily produces.

Iran’s economic condition can be genuinely terrible while its leadership simultaneously concludes that surrendering now would be worse.

And Trump just doesn’t get that.

That distinction is crucial.

Yes, Iran is suffering. But the question they seem to be asking is — is continuing to suffer a better option than accepting the terms currently available?

Looks like they think the answer to that one is yes.

And Tehran appears to believe that it still possesses something extremely valuable: the ability to impose continuing costs on Trump, the US, and the world.

Hormuz gives them that power.

Havng that leverage eliminates gives them the ability to impose costs that level the playing field and make them feel they have the upper hand.

Iran Is Watching Another Clock

There is also increasingly strong evidence that Tehran is paying very close attention to American politics, as well they should.

The Guardian’s Patrick Wintour reported last week that Iranian officials are openly studying the approaching November midterm elections and discussing Trump’s political vulnerability. Deputy Foreign Minister Kazem Gharibabadi has described a negotiating sequence that could take another two to four months — a timetable that conveniently runs directly into the American elections.

Two Very Different Readings of the Same War

And that brings us back to the apparent contradiction.

Trump says Iran is badly wounded and wants out.

Iran says America must first meet an extraordinary list of demands.

Iran almost certainly does want the war to end. There is little reason to believe its leaders enjoy watching their economy deteriorate and their military infrastructure being destroyed.

But wanting a war to end is not the same thing as being willing to end it on the enemy’s terms.

What Iran appears to be signaling is that it believes it has survived the phase in which overwhelming American airpower was supposed to compel capitulation. It has absorbed tremendous punishment and still retains the capability that matters most for coercing Washington: the capacity to interfere with regional energy flows and impose costs far beyond Iran itself.

Trump believes continued pressure will eventually break that resistance.

Iran appears to believe continued resistance will eventually break Trump’s willingness to maintain the pressure.

That may be the most important thing to understand about where this war stands today.

The battle is no longer primarily over who can inflict more physical destruction. There isn’t much question about that; the United States can.

The battle is over who can endure the consequences longer.

And right now, both Washington and Tehran appear convinced the answer is: we can.

<https://michaeldsellers.substack.com/p/iran-update-trump-thinks-iran-is> <https://michaeldsellers.substack.com/>

DEEPER LOOK with Michael Sellers

Iran update: Trump Thinks Iran Is Ready to Deal. Iran Thinks Trump is. Who’s Right?

If you listen to President Trump, the endgame with Iran sounds almost like a done deal. For more than a week now, Trump has been describing an Iran that is under enormous pressure, knows it is losing, and badly wants a way out. I know, how many times have we heard this. But each time we are compelled to do a reality check…

Read more

2 days ago · 94 likes · 23 comments · Michael Sellers

DEEPER LOOK with Michael Sellers

Former CIA officer, Russia specialist currently an investigator, taking a deeper look at things that matter.

Brad DeLong here: What are the important takeaways here?

It is wrong to even think that Iran might be bluffing. That is not a way to characterize it. When you face an untrustworthy adversary, there is no effective way you can negotiate a surrender that ends in a deal. You face a Clown Darth Vader: “This deal is getting worse all the time!” “I am altering the deal. Pray I do not alter it any further.” Only real Darth Vader changed the deal for terms you could understand, and foresee, because they made sense.

Let me briefly expand this, and then throw out an idea or two:

Read more

Successfully coordinating our lives, so far, we have a great many systems that together transform us from scattered jumped-up monkeys into the real ASI, the real Anthology Super-Intelligence that runs the world, the East African Plains Ape, coordinated in production and cognition. But we need metacognition as well: not just to think and do as a group, but to think as a group about how we think and do as a group. And of all the giants whose shoulders we find ourselves standing on, I do believe the tallest is Karl Polanyi, author of The Great Transformation.

Over on the really bad neofascist racist social-media website, Averroes writes:

Averroes: @delong thanks for introducing me to Polanyi.

I paid my respects to him today in Pickering, Ontario, by leaving him 20 Bulgarian lev…

Give a gift subscription

Karl Polanyi is the social theorist I would give the highest odds on with respect to their relevance to the world 50 years from now, the world of 2076.

If you asked me to name a number two, it would be John Maynard Keynes. But Polanyi definitely above Keynes.

And if you ask me to name a third, to try for a trifecta?

That is very very hard for me.

Let me give you a list. Why do these people make the list? Because each is, like Polanyi and Keynes, a master of one of the modes of large-scale coordination of the East African Plains Ape. For Keynes it is the market as a system. For Polanyi it is also the market as a system—but where the market as a system is always trying to destructively overreach, rather than how it works. But there are many more such systems pulling us together into the real ASI, not the Artificial but the Anthology Super-Intelligence of the East African Plains Ape. In rough descending order of the odds I would give each of still being genuinely useful for understanding the human world of 2076. :

1. Max Weber. The master-understander of bureaucracy. Plu: “charisma”, “legitimate domination”, and “the disenchantment of the world” are not going anywhere. As long as human beings organize themselves into large routinized and hierarchical structures and then argue about whose authority is rightful, Weber will be indispensable. He is my strongest third.

2. Alexis de Tocqueville. The master-understander of democracy. The great diagnostician of democratic society—of how equality of condition reshapes the soul, breeds both mass conformity and voluntary association, and lives in permanent tension with liberty. If the twenty-first century is about whether democracies can hold themselves together, he is the man to read.

3. Joseph Schumpeter. Technology meets market. Creative destruction is the closest thing economics has to a law of motion for a technological civilization. In a world about to be turned over by machine intelligence, the theorist of how innovation dissolves and reconstitutes whole social orders only gets more relevant.

4. Friedrich Hayek. Market, again. The problem of how dispersed knowledge gets coordinated without a central planner is a real and permanent problem, and it becomes sharper, not duller, in a world of algorithms. You cannot think clearly about markets without arguing with Hayek.

5. Hannah Arendt. Ideology. On totalitarianism, on the banality of evil, on the fragility of the space of public action—she wrote the field guide for how decent political orders decompose. I fear her century is not over.

6. Karl Marx. Political economy. Wrong about the coming revolution, wrong about the labor theory of value, wrong about a great deal—and yet nobody has ever been better on how the forces of production reshape everything else, or on capital’s restless compulsion to remake the world in its image. For thinking about the mode-of-production shift the machines are bringing, he remains a live wire.

7. Alexander Gerschenkron. Globalization. The theorists of how the periphery catches up—of the advantages of backwardness, and of how globalization unbundles and then re-bundles production across the planet. As economic gravity keeps shifting east and south, this is the frame for the world’s actual growth story.

8. W.E.B. Du Bois. Ethnicity. The color line was the problem of the twentieth century, and I see no evidence that hierarchy sorted by ascriptive identity is finished with us. Du Bois understood, better than almost anyone, how domination hides inside the categories a society treats as natural.

9. Michel Foucault. Algorithmic classification. I resist him, and I distrust the prose, but on how power works through knowledge, classification, and self-surveillance he anticipated the world of the database and the score. In an age of algorithmic sorting, the theorist of discipline and the panopticon has an uncomfortable claim on us.

10. Émile Durkheim. Imagination. Anomie, solidarity, the sacred, and the question of what holds a society together once the old bonds dissolve—these are exactly the questions a hyper-wealthy, hyper-connected, spiritually unmoored civilization keeps failing to answer. He is my dark-horse tenth: least fashionable, quietly essential.

But you ask, how about Smith, Machiavelli, Douglass—or Engels, Freud, Luxemburg, de Beauvoir, Gellner, Lewis or any of many, many others? Just below the cutoff.

Get 75% off a group subscription

Subscribe now

Leave a comment

If reading this gets you Value Above Replacement, then become a free subscriber to this newsletter. And forward it! And if your VAR from this newsletter is in the three digits or more each year, please become a paid subscriber! I am trying to make you readers—and myself—smarter. Please tell me if I succeed, or how I fail…

##the-grave-of-karl-polanyi-picture-of-the-day
##public-reason
##picture-of-the-day
##the-grave-of-karl-polanyi
##on-analyzing-market-bureaucracy-democracy-ideology-ethnicity-and-all-the-other-fictions-and-systems-we-east-african-plains-apes-have-imagined-into-reality-to-assist-us-in-coordinating-our-lives
#karl-polanyi
#john-maynard-keynes
#social-theory-relevant-for-2026
#east-african-plains-apes
#the-real-asi
#anthology-super-intelligence