A man sits on the floor of a study house in Jewish Babylonia, sometime in the third century, and teaches the same sentence for the four hundredth time. There are no books in the room. A text is a thing one human voice puts into another human’s memory and nowhere else, and the man, Perida, has a student whose memory will not hold it. So he says it again. The heat comes off the mud brick. His voice has gone to gravel. The student’s face is the same blank it was two hundred passes ago. He says it again.

Then he is called away mid-lesson, and when he returns the student has lost even the little he had, and Perida sits back down and starts from the beginning and teaches the same sentence four hundred more times.

The sentence never changes. Eight hundred identical passes at one fixed thing, and the loss function never goes down. The Talmud records this in Eruvin 54b, and it records the reward, which is the strange part. A voice from heaven offers Perida his choice: four hundred added years of life (I know what

@bryan_johnson

is picking), or the world to come for his entire generation (a new consideration for EA grants

@SBF_FTX

?)

Perida is not offered a faster student. Or a cleverer one. He is not offered a better lesson. The optimization target everyone assumes he was working toward is the one thing never put on the table.

I have published an essay every week for seven years. Now, my output is more than seven hundred of Substack essays (thanks

@hamishmckenzie

). I started by thinking I had built what the productivity literature calls a “deliberate practice” machine. But now I think I’ve built a different machine.

Anders Ericsson’s research is the serious version of the idea Gladwell flattened into “ten thousand hours.” Expertise comes from a particular kind of effortful, error-corrected practice rather than raw exposure. The violinist works the four bars she keeps missing, against the recording, until the deviation is gone. The chess player studies the lost position until the refutation is visible. Ericsson states a scope condition that the popularizers drop and that almost no one carries forward. Deliberate practice requires a known target and a fast, legible error signal. The passage has a correct sound. The position is winning or it is not.

But this familiar canon of skill acquisition examples (violin, chess) excludes precisely the work with the highest return, and it excludes it by definition, because there is no target to converge on and no signal to converge by.

According to Hayek, the reason central planning fails is not insufficient effort or bad planners. It is that the knowledge required to specify the optimal outcome does not exist prior to the activity that generates it. The target is not hidden from the planner; it is absent until the process produces it. It’s no coincidence that the people who push deliberate practice are new leftists. If only the government could just do more deliberate planning, everything would work.

But if the objective is absent until the activity produces it, then any system that operates by measuring distance to a specified objective is structurally barred from the activity — not slowed at it, barred from it — because it has nothing to measure distance to. The only way to learn is to FAFO, to just do things, to get feedback that changes not just the means but the ends. To improvise.

Deliberate practice is the planning model imported into the skull: assume a specifiable objective, measure the error, optimize. It works wherever the objective genuinely is specifiable, which is violin and chess and the things that look like them. It fails wherever the objective is discoverable only by the doing, for the identical reason the plan fails, and the failure is not a matter of degree. You cannot optimize toward a target that the optimization is supposed to produce.

Rabbi Yitzhak says (Megillah 6b) if a man tells you he toiled and did not find, do not believe him; if he tells you he found without toiling, do not believe him; if he tells you he toiled and found, believe him. Then the Gemara cuts the rule in half. This holds for words of Torah. In business, it says, success depends on mazal, on help from heaven. That cut is the calculation problem stated thirteen centuries early. Business is the domain of specifiable outcomes (John Doerr’s measure what matters; Andy Grove’s High Ouput Management), and there the effort converts in a linear manner.

But Torah is defined as the domain where the finding is not a result you reach but a state the toil leaves you in. Your nature itself is the product.

What did the repetition do, then, if it was not converging on anything. Chagigah 9b teaches:

The rabbis compare a man who reviews his learning a hundred times to one who reviews it a hundred and one, and they say the gap between them is the gap between one who has served God and one who has not.

According to options theory, this might be the difference, in practice, between someone who holds the option and one who sells it prematurely.

But the Talmud does not say the hundred-and-first pass corrects the material!

The material is not the variable. The man is.

The extra repetition does not improve the output by one percent; it moves the person across an inflection point. Just keep going.

And here is the part that should interest anyone who has watched a metric get gamed: the transformation is available only as long as it is not the objective. The moment the rep is performed in order to cross the line, it is instrumental.

Forty years old and unable to read, Akiva stood at a well and saw where the drip had cut clean through the stone (Avot d’Rabbi Natan 6:2). He asked who carved it. No one, they told him. The water did, the same drop, the same place, longer than anyone had counted. There is no version where he soaks the stone in a year. There is no flood. There is the drop and the same place and the not stopping, and a hole that when it finally goes through is total and was invisible the entire time it formed.

The consensus view in Silicon Valley is that general methods which scale with compute and feedback win, and that the lesson is bitter because it keeps being true. It is true, and its scope condition is the one Ericsson stated and the one Hayek stated and the one Megillah stated, which is that it holds wherever the objective is specifiable.

A model is the purest “shelo lishma” (extrinsically motivated) engine ever built. Every output is for a target. This is not a claim that the artifacts will be bad. They will be excellent. It is a claim that excellence of artifact and formation of person are distinct, and that the second was always the scarce thing.

So I do not wait to feel ready to write.

Readiness is a question about a target and there is no target. Friday morning comes. The essay that week may be the one I trip over. But it may also be the one that exists so the next can stand on it, the difference between serving the divine and not. I never know which, and the not knowing is the condition itself.

Perida sat back down and began again from the beginning, the eight hundredth time, no sign the student would ever have it.

I don’t write to express my thoughts. I write to discover the goal of my writing.

Or as Cavafy says, “Ithaka gave you the marvelous journey. Without her you wouldn’t have set out.”

- Zohar

P.S. - if you enjoyed this piece, you might enjoy Alexandria - our great books library with an AI tutor in the margins.