A four-word book review

Let's talk about the most important four-word book review in recent memory.

Marc Andreessen, cofounder of a16z, manager of tens of billions in tech investments, a man who has shaped how tech thinks for three decades, read Tyler Cowen's new book 𝗮𝗯𝗼𝘂𝘁 𝘁𝗵𝗲 𝗱𝗲𝗮𝘁𝗵 𝗼𝗳 𝗲𝗰𝗼𝗻𝗼𝗺𝗶𝗰 𝗿𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 and responded: "Self-recommending. Subscribe now!"

Four words. Subscribe now. Moving on.

I read it and opened another Claude tab. As one does

𝗕𝘂𝘁 𝘁𝗵𝗶𝘀 𝗯𝗼𝗼𝗸 𝗱𝗲𝘀𝗲𝗿𝘃𝗲𝗱 𝗺𝗼𝗿𝗲 𝘁𝗵𝗮𝗻 𝗳𝗼𝘂𝗿 𝘄𝗼𝗿𝗱𝘀, because what Cowen is actually saying is: we're automating our ability to know when AI is wrong about the economy, and calling it progress.

William Stanley Jevons, 𝗶𝗻 𝟭𝟴𝟳𝟭, didn't just invent marginalism, the framework underlying all of modern economics. He built the empirical program inside it: statistics, measurement, data. That program attracted smart people who did work that didn't need the framework. Then machine learning arrived.

𝗔 𝟮𝟬𝟮𝟰 𝗽𝗮𝗽𝗲𝗿 𝗶𝗻 𝘁𝗵𝗲 𝗝𝗼𝘂𝗿𝗻𝗮𝗹 𝗼𝗳 𝗙𝗶𝗻𝗮𝗻𝗰𝗶𝗮𝗹 𝗘𝗰𝗼𝗻𝗼𝗺𝗶𝗰𝘀 pitted a 360,000-factor model against the traditional 6-factor model. The 360,000-factor version won. It cut pricing errors by 54.8 percent. Nobody can explain what 359,994 of those factors mean. The field published a paper, in its own house, saying it no longer needs the house.

Steve Levitt quit academic economics at 57. 𝗙𝗿𝗲𝗮𝗸𝗼𝗻𝗼𝗺𝗶𝗰𝘀 𝗦𝘁𝗲𝘃𝗲 𝗟𝗲𝘃𝗶𝘁𝘁, the man who made economics readable to people who don't voluntarily read economics, said on his way out: "Chicago price theory has essentially lost. Lost to posterity at this point."

𝗠𝗶𝗹𝘁𝗼𝗻 𝗙𝗿𝗶𝗲𝗱𝗺𝗮𝗻 said the same thing in the 1990s. Nobody listened.

𝗧𝗵𝗲 𝗨𝗻𝗶𝘃𝗲𝗿𝘀𝗶𝘁𝘆 𝗼𝗳 𝗖𝗵𝗶𝗰𝗮𝗴𝗼 now runs a summer camp. The Price Theory Summer Camp, "created to introduce PhD students from outside the University of Chicago to price theory." PhD students. Summer camp. In the subject that used to be the core curriculum of a PhD.

𝗔 𝘁𝗲𝗻𝘂𝗿𝗲𝗱 𝗛𝗮𝗿𝘃𝗮𝗿𝗱 𝗲𝗰𝗼𝗻𝗼𝗺𝗶𝗰 theorist posted publicly that "economic intuition" now means "math introduced to economics more than 20 years ago." A senior economist at another major program replied: "Intuition, Shmintuition, look at the proof."

When GPS fails, in a tunnel or after a solar event or during signal jamming, the sailor who never learned celestial navigation has no backup.

We are producing a generation of policy analysts who will inherit 360,000-factor models they cannot audit, making decisions affecting people who never chose the model, in domains where being wrong doesn't show up in a quarterly report.

𝗖𝗼𝘄𝗲𝗻'𝘀 𝗹𝗮𝘀𝘁 𝗹𝗶𝗻𝗲: "Maybe our intuitions about the world were never so strong in the first place... swimming in a larger froth of epistemic chaos. And now the illusion has been stripped bare."

Andreessen read that.

I read it and opened another Claude tab.

Nobody's building the backup. Right now, it's getting four words.