@patio11
Post
Post 1 of 10
The good news: we got much better at running The Sort. The bad news: we got much better at running The Sort. Quoted post by Crémieux (@cremieuxrecueil) The dollar value of an additional IQ point, by age, in two American cohorts about a generation apart: 
Image from X post Open quoted post on X
Post 2 of 10
We, the societal we, implemented a relatively effective nationwide (and increasingly worldwide) dragnet for talent, then plucked that talent from the not-random-but-constrained walk through lives it would counterfactually have encountered and tracked it fairly narrowly.
Post 3 of 10
The Sort makes many mistakes, in both directions. It leaks promising candidates. The Sort doesn’t really feel good or bad about that, because it is an emergent behavior of a system comprised of many disparate actors with differing incentives, values, and similar.
Post 4 of 10
But, if the Sort would tolerate anthropomorphization (knowing which word it would award you points for, not that many, just the properly calibrated amount), it would look at many outcomes and declare them inefficient.
Post 5 of 10
The Department of Public Works in a small Iowa town is a competent administrator? How competent? Because most forms of competent are inefficiently competent. The Sort would prefer him to have ended up with two OOM more budget and four OOM more impacted users.
Post 6 of 10
The Sort, hearing family history, thinks that it was an obvious mistake that my father was a paralegal (swap with worst lawyer, Pareto improvement) and thinks it got closer with regards to me and still squandered a decade. Not a value judgement; just inefficient, you understand.
Post 7 of 10
The Sort is incredibly aware of agglomeration effects. If one doesn’t understand them, NP, low-productivity areas also require labor forces. One will enjoy taking in the cultural delights of the world’s foremost cities when one can afford to vacation to them.
Post 8 of 10
(The Sort didn’t ban building housing in those cities. It is as puzzled by that as everyone else is. It suggests it’s usual remedy: maybe if we just sort harder the problem will go away. The Sort is correct in its perception that this has worked in many contexts for it.)
Post 9 of 10
The Sort is incredibly aware of criticisms of the social implications of the Sort, and has devoted an appropriate number of 90-95th percentile sorted to thinking about that important, but not very important, problem.
Post 10 of 10
The Sort is quite aware that many people viscerally hate descriptions of its operations. It has decided to fib about them as a stakeholder management strategy. Capacity with encoding and decoding this sort of fibs is sorted for, naturally. They’re really quite efficient.
Explanation
Patrick McKenzie’s point is that modern America has become much better at identifying cognitively talented people and routing them into a relatively narrow set of high-productivity institutions, occupations, and cities. He calls this emergent system “The Sort”: not a conspiracy or a single bureaucracy, but the combined effect of schools, standardized tests, selective universities, employers, recruiting, professional credentialing, geographic migration, and social networks. The opening line — “The good news: we got much better at running The Sort. The bad news: we got much better at running The Sort.” — means that this process is economically productive but also socially consequential.
The chart he is reacting to compares two US longitudinal cohorts, the NLSY79 and NLSY97. These are large government studies following people born roughly a generation apart: 1957–64 versus 1980–84. ([National Longitudinal Surveys][1]) The plotted quantity is essentially the estimated extra annual income associated with one additional IQ point, at different ages. The striking feature is not really that the younger cohort gets a dramatically higher return to IQ at the same age—the underlying research actually found the age-20-to-29 returns fairly similar across the two cohorts—but that the income payoff to cognitive ability rises strongly as careers progress. At age 28, for example, a 0.1-SD increase in AFQT cognitive score was associated with about 2.0% more income in the later cohort versus 1.7% in the earlier one, a statistically insignificant difference. ([National Bureau of Economic Research][2]) So McKenzie is using the image somewhat more broadly than “this graph proves sorting intensified by a generation.” His argument is an interpretation of the surrounding social machinery, not a result established directly by that graph.
“Plucked that talent from the not-random-but-constrained walk through lives it would counterfactually have encountered” means that in an earlier or less efficient system, a very able person might simply become the unusually competent manager of a small-town public-works department, local accountant, technician, teacher, etc., largely because of birthplace, family occupation, local contacts, or limited information. The modern system increasingly detects that person and moves them toward Stanford/MIT, Google, finance, elite law, medicine, consulting, startups, New York/SF/London, and so forth. McKenzie’s Iowa administrator example is deliberately cold: from the system’s perspective, someone capable of administering a $2 million operation serving 20,000 people might be “wasted” if the same person could administer a $200 million operation affecting 200 million people. His “two OOM” and “four OOM” mean two and four orders of magnitude: roughly 100× more budget and 10,000× more users.
“Agglomeration effects” are why the Sort also pushes talent geographically. Highly productive workers often become even more productive when clustered with other highly productive workers, firms, capital, specialized suppliers, and dense labor markets. So New York, London, Silicon Valley, etc. pull in disproportionate talent. His housing aside is a joke about the resulting contradiction: the economy wants more talented people in productive cities, while zoning and housing restrictions make living there extremely expensive.
The final jokes are darker. McKenzie says the system is sufficiently sophisticated that it also recruits smart people to study and manage the social problems caused by sorting—but, in his caricature, only as much as is instrumentally useful. And because explicit descriptions like “we identify unusually capable children and channel them toward elite institutions where their abilities earn much more” sound meritocratic, elitist, or dystopian depending on the listener, institutions often describe the process in softer language about opportunity, fit, leadership, access, holistic evaluation, and so on. His punchline is that learning to fluently produce and decode those euphemisms is itself another ability that elite institutions select for.
The important caveat is that McKenzie is not claiming The Sort is omniscient. He explicitly says it misses talent in both directions. His thesis is that it has nevertheless become powerful enough that the remaining “misallocations”—a brilliant paralegal, an exceptional local administrator, a gifted person who never enters the elite pipeline—start to look to the system like economic inefficiencies rather than just ordinary variation in human lives.
[1]: https://www.nlsinfo.org/content/cohorts?utm_source=chatgpt.com "Cohorts | National Longitudinal Surveys" [2]: https://www.nber.org/system/files/working_papers/w22470/w22470.pdf?utm_source=chatgpt.com "We next estimate models for 20-29 year olds using common specifications and controls using data from the NLSY79 and NLSY97 cohorts. The results, summarized in Figure 3, suggest that the returns to cognitive performance are extremely similar across the two cohorts across all of the ages studied, with some possibility of a (statistically insignificantly) higher return for NLSY97 respondents. For example, at age 28, the point estimate (standard error) of the increased income associated with a 0.1 standard deviation rise in AFQT scores was 2.0 (0.2) percent for the later cohort versus 1.7 (0.1) percent for the earlier one."