How to Learn

This page is where I keep learning systems I find interesting enough to try: methods for getting something into your head, understanding it fairly, and turning it into something of your own. Each one gets a section, and the page grows as I find more.

The genius curriculum

Six civilisations, six unrelated educational systems, each one producing great thinkers at a rate no modern university comes close to. The claim in the video is that they were doing six different things that share a hidden set of features, and that you can reassemble those features into a curriculum for learning practically anything.

These are my notes on that argument, plus the five-session loop it ends on. Four of the six only make sense once you’ve handled them, so those have something to poke at.

1 · The permutation

The Vedic tradition had to move an enormous body of text across generations with every syllable intact, and without writing any of it down. Whisper Down the Lane says that should have been impossible.

The fix was to stop treating recitation as a sequence. Students learned the text in as many as eleven permuted patterns — in jaṭāpāṭha the words are braided together and recited back and forth. Someone who genuinely knows the material can produce it in any of the eleven forms. Someone who has only memorised the order falls apart the moment the order is scrambled.

Here are the classical patterns, generated from whatever you type. Start at pada and walk right; watch the counter under each word.

It worked. The Rigveda came down roughly three millennia of oral transmission as essentially one text, with a fidelity no other ancient corpus matches. And the system produced Pāṇini, who around the 4th century BCE composed the Aṣṭādhyāyī — a complete generative grammar of Sanskrit in about 4,000 rules, which Leonard Bloomfield called one of the greatest monuments of human intelligence and which computer scientists now compare to a Turing machine. He never wrote it down. He built the whole thing inside the oral system.

The principle: you have not learned something until you can produce it in an order other than the one you learned it in. Forwards, then backwards, then sideways — and wherever the versions disagree, you have just located the exact place your understanding is broken.

2 · The empty mind

Zhu Xi, in 12th-century China, left behind something almost nobody else bothered to write: an explicit methodology of reading. Two of its principles do most of the work.

The first is hányǒng — empty the mind and swim in it. Before you evaluate a text at all, suspend your own opinions entirely and let the thing speak. The second is the opposite move, and it comes second on purpose: set yourself against the text personally and ask what it demands of you.

Almost nobody actually does the first one. Here is a five-sentence argument and a switch. Read it in the state you are probably already in, then flip it.

Zhu Xi’s insistence was that comprehension and evaluation are two separate activities, and doing them simultaneously wrecks both.

In 1313 the Chinese state made his commentaries on the Four Books the official basis of the imperial examination system. They held that position until 1905 — nearly six centuries — while also shaping education in Korea, Japan and Vietnam. Essentially every credentialed thinker in East Asia for six hundred years came up through this reading method.

3 · The chain of custody

From the Islamic scholarly tradition, and specifically the discipline built around hadith: the isnād. A claim does not travel alone. It travels with its chain — the documented sequence of who heard it from whom, all the way back.

The institutional version was the ijāza, a licence to transmit a specific text, granted by a teacher who himself held one. In principle you could only teach what you could trace.

Pull on one and watch where the numbers come from.

The example is al-Bukhārī, working in the 9th century: sixteen years travelling across the Islamic world, over a thousand teachers, something on the order of 600,000 narrations examined, of which he retained around 7,000.

The principle: take the single claim your whole position rests on and trace it back. Who says it, citing whom, citing what, until you hit a primary source.

4 · The quaestio

Around 1121 Peter Abelard produced Sic et Non — “Yes and No” — laying out 158 theological questions and, for each, assembling the authorities that flatly contradict each other. Then he supplied no resolutions at all.

Out of that grew the quaestio disputata, which Aquinas formalised into what is still one of the most rigorous thinking templates ever devised. Every article in the Summa runs the same way:

1. The objections — the strongest arguments against the position.
2. The sed contra — a single consideration on the other side.
3. The response — his actual answer.
4. The replies — each objection answered individually, in turn.

Look closely at that ordering, because the ordering is the whole thing: you are required to state the case against yourself before you are permitted to state your own. So here it is with the ordering actually enforced. Bring a real question — something you already think you have settled.

Descartes spent eight years at the Jesuit Collège de La Flèche on a curriculum built on Aristotle and the Summa, trained in disputation under aemulatio — rivalrous play in pursuit of excellence. He then spent his career dismantling scholastic philosophy. But look at how he published: the Meditations came out with six sets of Objections and Replies appended, solicited from other philosophers. He overthrew the content using its own format. The doctrine died; the method is still one of the best tools we have.

5 · The ḥavruta

The quaestio has an obvious limit: you cannot really be your own opponent.

In the yeshiva tradition the primary mode of study isn’t lecture and isn’t solitary reading. It’s ḥavruta — paired study, out loud, for hours, where your partner’s job is to attack you. The Talmud grounds it in Proverbs: iron sharpens iron, and so two scholars sharpen each other.

It inverts almost everything modern professional life trains you to do. Agreement is the sign of weakness.

6 · The unsolved problem

Everything above handles ideas that already exist. None of it makes you produce anything new.

At the Fasori Lutheran Gimnázium in Budapest, a mathematics teacher named László Rátz taught both John von Neumann and Eugene Wigner. The wider Budapest system of that era also produced Leó Szilárd, Edward Teller, Theodore von Kármán, George Pólya, Paul Erdős and John Harsányi.

Rátz inverted the normal order of instruction: students were put in front of hard problems before they were taught the technique that solves them, and were expected to construct the approach themselves. Running alongside it was KöMaL, a journal founded in 1894 by Dániel Arany and edited by Rátz from 1896. Each month it posed deep problems with deliberately no time pressure — so the problems could be genuinely difficult rather than merely fast. Teenagers did original work in public, under their own names, read by their peers across the country.

So: a problem, and no technique.

The five-session loop

Choose one thing: one paper, one chapter, one problem, one argument. Spend five sessions on it across two or three weeks. If that feels absurdly slow, that reaction is precisely the thing all six traditions were built to overcome.

Session 1 · The fair statement — 40 minutes

Read it once. Reread it under a single prohibition: you may not evaluate. No disagreement, no objections in the margin. Then write one paragraph stating the author’s case so well that the author would approve of it. Hand that paragraph to someone who knows the material, or check it against the text line by line.

Session 2 · The scramble — 45 minutes

Close the source. Fifteen minutes: write a forward summary from memory. Fifteen minutes: write it backwards — conclusion first, then what it rests on, all the way back to the opening premise. Then name the three central claims and explain each one in terms of the other two.

Session 3 · The warrant — 30 minutes

Take the one claim the whole thing rests on and trace it back. Who asserts it, citing whom, citing what, until you reach a primary source. Write the chain out. Then run two tests at every link: is this source reliable, and what can I — and can I not — actually conclude from it?

Session 4 · The case against — 50 minutes

State the question as a strict yes or no. Then, before writing one word of your own answer, spend twenty minutes writing your three strongest objections to your own position, strongest first. Add the single best consideration on your side. Then your answer, in one paragraph. Then reply to each objection individually, by number.

Any objection you cannot answer now becomes your actual position.

Session 5 · The live test — 45 minutes

Out loud, with one partner. You state your answer; for twenty minutes they give objections only. Then swap, and argue the position you just spent twenty minutes demolishing. Use the last five minutes to write down what changed.

Within 48 hours

Two external checks, because either one alone will miss something:

• Report your conclusion to a third person in under three minutes.
• Publish your written attempt somewhere public, with your name on it, dead ends and mistakes included.

Teaching one obvious step at a time

Eero Alvar built an open-source desktop app that teaches you with an AI agent, and his video on it is the clearest account I’ve found of how teaching works. I don’t use the app. I took the process, because it’s the one I try to follow when I write an explainer, and he put it into words better than I had.

His goal is the most understanding for each unit of mental effort. Struggling with the material is where the learning happens, so the aim is to spend your effort on the material itself and as little as possible on everything around it. His biggest worry with AI is that it lets you skip the struggle and leaves you with an illusion of learning.

Understanding isn’t recall

You can follow a lecture, understand every step, and never think of the idea again when a problem needs it. The understanding was real. Recognising when to use it is a separate skill. Recall runs on cues, the features of a situation that bring an idea to mind, and a new idea starts out tied to the room it was taught in.

So he splits learning in two. Teaching puts the understanding in, and practice widens the set of cues that can bring it back.

Every step obviously true

A step is only redundant to someone who already knows it, and only out of reach to someone missing what it rests on. So his agent starts by probing, with quick multiple-choice questions that find the edge of what you know. Then it plans the shortest chain of ideas from that edge to the goal.

Then it walks the chain, and every step has to land as obviously true given the one before. That’s achievable because each step is a single move. Keep going and the goal itself turns into one more obvious step. It’s the best line I’ve heard for what I’m after when I write, and it’s now a rule in my own writing guide.

Each step has the same shape: predict, teach, check. Before anything is explained, you guess the next move. The guess shows how you’re thinking, and reasoning it out yourself is learning in its own right. A wrong guess surfaces the confusion before anything gets built on top of it.

Thinking out loud

A right or wrong answer says very little on its own, since two people can give the same wrong answer for different reasons. So nearly every screen in his app has a box for your raw, messy thinking, and that’s the main thing the agent reads. It’s also why he keeps every step short. Short steps mean the agent hears from you constantly, so each explanation fits what you actually think.

Recall that takes effort

In practice, the agent’s job is to make you recall the right thing from a set of cues. The problems adapt as you go, aiming for recall that succeeds with effort: hard enough to make you work, easy enough that you get there. They get harder by combining more ideas or by making the cues less familiar. When you’re stuck you get the smallest nudge that still lets you find the answer yourself.

Linking the two halves is a record of everything you’ve been taught, kept as short statements and the reasoning between them. It draws as a graph that grows with each session, and it’s what lets the next lesson find your edge quickly.

The loop without the app

None of this needs his software.

• Before reading an explanation, write down what you think comes next.
• Work problems with your reasoning out loud or on paper, mess included.
• Once you’ve learned something, use it on a problem that looks nothing like the example it came with.
• When a step doesn’t feel obvious, the gap is in the step before it. Go back one.

Sources

Everything above is a synthesis of two videos, not independent research. The learning-harness section describes Eero Alvar’s own system and his reasons for it, watched in November 2026; I haven’t tested his claims about what works. In the genius-curriculum section, the names and figures are the ones that video asserts; I have normalised the transliterations (the auto-captions mangle Zhu Xi, hányǒng and ḥavruta) but I have not gone back to the primary sources to check the claims, which is exactly what Session 3 would have me do.