Superlearning

December 3rd, 2024

“What is it you do to practice that is analogous to how a pianist practices scales?” - Tyler Cowen

Over the past 6 months, I’ve entirely overhauled how I learn. Previously, my system for learning was fairly unstructured. I aimed to read a book per week, with some other short-form content sprinkled in. I would either read what was interesting to me, or something that was very different from what I’d previously read.

Beyond that, I didn’t have much of a process. I didn’t write book reviews, and I rarely had a reading group to discuss the material with. But I wanted to track my educational progress, and without anything else to grab onto, I started tracking the number of books I read.

Over time, I became increasingly dissatisfied with this measure of learning. For one, I began noticing that I was forgetting almost everything that I read; despite all the time I spent, it wasn’t clear that my return on investment was high. And second, because the number of books I read was my measure of progress, it unintentionally started to become my target.

The more it became my target, the more my learning deteriorated, a la Goodhart’s Law. Racing to finish each book, I would skip sections, pass over concepts that I didn’t understand, and hasten to pick up the next book before fully digesting the previous one.

I needed something better.

A Better Solution

What would a better system look like?

Well, my goal is to become more intelligent. I think much of intelligence – at least the portion of intelligence that we can improve – is just association, that is, the ability to understand concepts and draw connections between those concepts. So to be more intelligent, we should aim to increase the number of concepts we understand and can easily reason across.

Therefore, a better system of learning is to maximize the amount of useful information one has stored in long-term memory. Note I’m not saying that this is the perfect approach; I’m claiming that this is a much better approximation of what it means to “become smarter”.

New System

My new system is designed just for this.

It consists of three parts: exploration, atomization, and retention. At the foundation of the system is a technique called spaced-repetition learning. The technique is commonly used by medical students and other learners who must memorize vast quantities of information. It involves reviewing a piece of information at the optimal frequency to commit it to long-term memory as fast as possible.

To practice spaced repetition learning, I use a software program called Anki, which allows you to make flashcards and review them at the optimal point along the forgetting curve. After reviewing a card, the software lets you rate the material’s difficulty, and will then re-present the content at a certain time in the future based on that difficulty.

In my new system, instead of tracking the number of books read, my measure of progress is the number of Anki cards committed to long-term memory.

Let’s look at each step of my process – exploration, atomization, and retention – one-by-one.

Exploration

Exploration is simply the process of consuming new content. It’s reading, listening to a lecture, watching a film, or absorbing information in any form. I try to explore as much as possible and aim to go both deeper into familiar topics and wide into new topics.

Good exploration is hard to teach. It’s mostly about one’s taste in content, and the curiosity to pursue it.

Atomization

During exploration, if at any point I come across something that I don’t understand and/or want to remember, I stop, and atomize it.

Atomization is the process of taking a chunk of information, and breaking it down into single, standalone units. For example, consider these 2 approaches to new material:

Approach 1:

Who was Rembrandt?

Rembrandt was a Dutch Gold Age painter, printmaker, and draughtsman. He is generally considered one of the greatest visual artists in the history of Western art. It is estimated that Rembrandt produced a total of about three hundred paintings, three hundred etchings, and two thousand drawings.

Approach 2:

What country was Rembrandt from?

Netherlands

What important era of Dutch history did Rembrandt live during?

Dutch Golden Age

Approximately how many paintings did Rembrandt produce?

Three hundred

Clearly the 2nd approach is more atomized. Each question and answer contains just one idea, and is therefore much easier to memorize. While atomizing content takes more time, the extra work pays itself back during the memorization process, and makes reviewing content more enjoyable.

Here’s my algorithm for atomizing information quickly and effectively:

If the material is atomized and adequately explained where you’re currently reading it, simply move it over to an Anki card.
If it’s not, look up the material either using Google or Perplexity. I find for very simple cards, like definitions, Google is sufficient. But for more esoteric cases or situations where the content is hard to simply look up, Perplexity is superior.
Does the card have a good visual aid? If yes, find an image on Google, and copy it over. Visual aids make learning content much easier.

Some content is difficult to atomize, like philosophical ideas that are hard to summarize. For example, here is my card on Soren Kierkegaard’s notion of the Teleological Suspension of the Ethical.

I know that this doesn’t fully capture the nuance of his idea. But it’s enough to understand the basic intuition.

Another domain of knowledge that is difficult to atomize is complex systems like biology, where some things have many overlapping or partial effects. For example, it would be correct to say that “the Cerebellum controls balance for walking, standing, and other complex motor functions.” I could then make a card that says that that is what the Cerebellum is for, but it wouldn’t reflect the full range of the Cerebellum’s functions, like its importance for eye movement, language, and attention. If I try to make a card that lists all of the Cerebellum’s functions, it would violate the principle of atomization by containing too much content.

The solution in this case is to settle for approximate or partial truths, and to understand that there’s more detail than can fit on a short card. I think of this type of material as setting the foundational bricks upon which further cards can be created that add that additional content, if I ever want to go deeper. And since the foundational concepts are already committed to long-term memory, a more thorough review becomes much easier.

There are many techniques besides atomization that are important for making effective Anki cards. Check out the links at the bottom for more tips.

Retention

The last and most important part of my process is retention via reviewing Anki cards every day. This is about taking everything I’ve done so far, and making sure it doesn’t go to waste.

Retention is about consistency and diligence. Skipping days knocks one off of their optimal retention curves, increasing the time it takes to memorize content. Worse, missing days means that review cards pile up, creating a higher burden for getting back on track. It can be demotivating to stare at a mountain of 1,000 flashcards that need review.

But when done right, this technique allows you to dramatically expand your long-term memory and casually memorize knowledge across any discipline with just 20-30 minutes of review per day. Here’s a representative sample of 10 cards in my deck:

Conclusion

When you change your metric (and in effect your target), you change your behavior. I’ve noticed that since changing my metric for learning from the number of books read to the number of cards stored in long-term memory, my exploration patterns have also changed.

I spend less time forcing my way through long technical books. I rush through material less, and am a more active reviewer, since I’m always stopping to make Anki cards. I also spend more time with online material versus physical books, since they’re often faster to Ankify and more information dense.

Since my technical reading is more online, my remaining physical reading is more centered around literature and other content that isn’t as accessible in an online format.

This system of exploration, atomization, and retention has dramatically improved my learning velocity and effectiveness. It is, for me, a system for superlearning.