Active Learning — Keeping 70 Percent by Writing and Teaching

• Opening — I clearly read it, so why did none of it stick?
• Core insight — memory grows stronger when you pull it out
• The cognitive science of active learning — why retrieval changes encoding
  • The generation effect — an answer you make beats an answer you receive
• Going deeper 1 — writing is a debugger for thought
  • Concrete ways to learn by writing
• The real process of finishing one piece of writing (a step-by-step walkthrough)
• Going deeper 2 — the protégé effect: learning by teaching
  • When you have no one to teach
• How to teach well — the skill of imagining your audience
  • Overcoming the curse of knowledge
• Combining active learning with spaced learning
• The active-learning research is real, but those numbers are fake
• Comparison — passive vs. active learning
• Common excuses vs. rebuttals
• Practice — a routine that turns 70 percent of input into output
  • An example of input vs. output time allocation
  • An example weekly routine
• Pitfalls — what happens when you misread active learning
• FAQ
• Closing — let us do study that stays
• References

Opening — I clearly read it, so why did none of it stick?

Last year I read a book on distributed systems cover to cover. It was a thick book.

I plowed through it on the LINE office commuter shuttle every single day, and after two months I finally closed the last chapter. I felt proud when I finished. I had poured two months of morning commutes into that one book.

But a month later, when a colleague asked me, "Remind me how leader election works in consensus algorithms," I could not explain it properly. I had definitely read that chapter. I had nodded along while reading it. Yet when I tried to put it into words, it had scattered like fog.

That day shook me a little. Two months of effort, and this was all that remained? What stayed in my head was only the impression "that was a good book," while the actual contents of the book slipped right through my fingers.

Then a memory suddenly surfaced. Back when I was a junior, I had written and posted a deployment pipeline document on the internal wiki myself. Two years later, that one was still almost completely intact in my head.

What was the difference? One I simply read; the other I wrote out by hand and explained to others. And yet the time I had invested was far greater on the book's side.

The truth is, I had had many similar experiences before that. Times when I finished an online lecture course end to end but went blank in front of the exam paper. Times when I watched a conference talk in admiration, only to forget even its title a week later. Each time, I blamed my memory.

But looking back, the problem was not my memory; it was my method. I had always only "put things in" and never once "pulled them out." Knowledge that you only put in leaks back out — that is not a memory problem, it is the natural result.

This post is about that difference. The conclusion first: input alone (reading and listening) does not stick well, and only after passing through output (writing and teaching) does knowledge finally become yours.

That said, I will also point out that the clean numbers so often quoted — "the learning pyramid, active learning means 90 percent retention" — are largely exaggerated. The conclusion is right, but the numbers used as its evidence are frequently fake.

Core insight — memory grows stronger when you pull it out

Let me start with the firmest evidence. Cognitive psychology has a well-verified phenomenon called retrieval practice, or the testing effect.

The idea is that the very act of pulling what you learned out of memory, rather than rereading it, makes the memory stronger. The key is not "putting in" but "pulling out."

A representative study is the experiment Karpicke and Roediger published in Science in 2008. They had students memorize foreign-language words; one group kept rereading them, while another group kept testing themselves.

A week later, the result was dramatic. The group that tested itself remembered far more than the group that only reread. The key was the act of pulling the information out.

What is even more interesting is that the students' predictions were exactly the opposite. The students who only reread were confident they would remember better, because familiarity gave them confidence. But their actual scores were the reverse.

Why is that? The moment you strain to retrieve information from memory, your brain repaves the path leading to it. Just as a trail grows clearer the more you walk it, retrieval strengthens the route to a memory.

By contrast, simply rereading only gives the feeling of familiarity — "I know this" — without walking the trail again. And so we deceive ourselves often, mistaking familiarity for understanding.

This illusion even has a name: the fluency illusion. Because something reads smoothly, you feel you know it. But reading smoothly and being able to pull something out of a blank page are entirely different abilities.

Writing and teaching are, at their core, powerful acts of retrieval. When you write out an explanation on a blank page or explain something out loud to another person, you continuously pull from memory, discover the gaps, and fill them back in.

This is exactly why reading is passive while writing and teaching are active. Reading is gazing at the trail; writing and teaching are walking the trail yourself.

The cognitive science of active learning — why retrieval changes encoding

Let me go a little deeper. Once you understand the mechanical principle behind "pulling it out makes it stronger," it becomes natural to see why the practical methods are shaped the way they are.

Memory is usually described in three stages: encoding, storage, and retrieval. We commonly fuss only over the encoding stage — the "putting in" stage. We believe that if we read harder and look more times, things will go in better.

But the insight that retrieval practice gives us is different. Retrieval is not a passive lookup that merely reads back what was stored. The act of retrieval itself rewrites the memory and makes the next retrieval easier.

Cognitive psychologists describe this by saying that "retrieval is a powerful modifier of memory." Every time you pull it out, that memory hardens a little more. It is similar to how a muscle adapts every time you use it at the gym.

Another concept overlaps here: the levels-of-processing effect. The idea is that information processed shallowly stays shallowly, while information processed deeply stays deeply.

For example, looking at a word and judging "is this letter capitalized?" is shallow processing. By contrast, thinking "how does this word connect to my own experience?" is deep processing. The latter is remembered overwhelmingly better.

Here is yet another reason why writing and teaching are so powerful: both force you to process information deeply.

To write, you have to convert information into your own sentences. To convert it into your own sentences, you have to connect that information to something you already know. To teach, you go one step further and have to imagine the inside of your listener's head.

I often compare this process to data transformation. Reading is just loading data into memory. Writing is serializing that data into another format. If there is a part that will not serialize, then that data was actually broken in its structure all along.

This is the real meaning of "changing the encoding." When you study with retrieval as the premise, the very way you put things in changes. The moment you think "later I will have to write this out on a blank page," you naturally start reading by sifting out the core and building a structure.

So active learning is not only a story about the output stage. The moment you keep output in mind, the quality of your input changes along with it. This is the heart of the claim that retrieval changes encoding.

The generation effect — an answer you make beats an answer you receive

There is one more concept that pairs with the levels-of-processing effect: the generation effect.

It is the phenomenon that, for the same information, something you produce yourself is remembered better than something spoon-fed to you. Seeing the correct answer in a blank versus filling the blank in yourself produces different results, even with the same time spent.

I felt this acutely when learning to code. Reading someone else's code a hundred times mattered far less than covering it up and writing it once myself. Code that seemed completely understood while reading would not come out at all once I faced a blank screen.

That gap is exactly what the generation effect points to. Seeing is easy; making is hard. And the hard side is the one that stays.

Writing and teaching are both "generation." On a blank screen I produce the sentences myself; on a blank whiteboard I draw the diagram myself. In that they are acts of making rather than transcribing, these two are the most powerful generative activities.

To sum up, the power of active learning does not come from a single principle. Retrieval practice, levels of processing, and the generation effect all operate at once. When we write and teach, we enjoy all three of these at the same time.

Going deeper 1 — writing is a debugger for thought

Many people say, "I am bad at writing." But this comes from mistaking writing for "crafting pretty sentences."

As a learning tool, writing is not about style; it is about organizing thought. To use an analogy, it is closer to an athlete checking their own form in the mirror. The point is not to look impressive but to confirm where something has gone wrong.

A thought always seems plausible inside your head. Even as a vague blob, it gives the feeling "ah, I understood it."

But the moment you move it into sentences, the blob breaks apart and the holes reveal themselves. A point appears where you cannot answer the question, "So why is that the case?"

That point is exactly where you actually did not know. Writing is a debugger that automatically finds the empty spots in your understanding. A bug that you would never have seen if you just sat still reveals itself like a stack trace as you write line by line.

I compare this to code review. The thoughts in your head are code you have not compiled. The syntax seems fine and the logic seems fine, but you cannot know until you actually run it.

Writing is actually running that code. The line that throws an error — the spot where your explanation stalls — is exactly the spot you need to shore up.

An anecdote about the physicist Richard Feynman is often cited. When he met a difficult concept, he would try to build a lecture that even a freshman could follow; if he could not, he took it as a sign he did not yet understand it himself.

That is the heart of the so-called Feynman Technique: if you cannot explain it simply, you do not yet know it. Glossing over things with difficult jargon is often not evidence of understanding but a screen for the lack of it.

Concrete ways to learn by writing

• Blank-page summary after reading: Close the chapter and, without looking at the book, write the key points on a blank page. Mark where you stall, then reopen only those parts.
• Write "why" three times: Write down a conclusion, add "why?" beneath it, and write the answer. Add "why?" to that answer again. If you survive three levels, surface knowledge and real understanding part ways.
• Write something a reader will see: A note only you read and a piece a colleague will read have different precision. The moment you assume a reader, vagueness disappears. This blog is, in fact, one of my learning tools.
• Make one analogy: Liken the concept you learned to something in a completely different field, in a single sentence. If a good analogy comes out, you have properly grasped the structure of that concept.

The real process of finishing one piece of writing (a step-by-step walkthrough)

Just being told "organize it in writing" leaves you at a loss. So let me lay out, step by step, the process I actually go through to write one learning-oriented piece. The example topic is distributed locks, which I organized recently.

Step 1 — start from a blank page. Before opening any reference, I first write down as much as I know on a blank screen. "A distributed lock keeps multiple servers from touching the same resource at the same time" comes out smoothly. But at "so what happens if the server holding the lock dies?" my hand stops. This stopping is the first harvest.

Step 2 — fill only the spots where you stalled. I do not reread from the beginning. I look up only the place where I got stuck — "what to do when the lock holder dies." Timeouts and the concept of a lease come up. Reading only this part with focus, it goes in far better.

Step 3 — pick one reader. I usually take "myself from six months ago" as my reader. When I recall what that version of me did not know, it becomes clear where I should start explaining. The key is to imagine one concrete person, not a vague "someone."

Step 4 — structure first, sentences later. I do not cling to a pretty first sentence. First I just list the subheadings: "why it is needed / the problem with the simple approach / the lease-based solution / the remaining pitfalls." Once the skeleton stands, the flesh attaches quickly.

Step 5 — getting stuck is a key discovery. While writing, I got stuck on the question, "Why is it bad to set the lease time too short here?" The reference did not spell it out clearly either. This stall was precisely the real gap I had been missing, and I filled it by experimenting separately and asking a colleague.

Step 6 — read it aloud again. I read the finished piece out loud. The places that sound awkward when read aloud are usually the places where the logic is awkward. It is not the pronunciation that stalls but the understanding.

Of these six steps, the most important are steps 1 and 5. Both are stages where you actively hunt down the "stall." Writing that never stalls is actually closer to copying, and copying is not retrieval.

Going deeper 2 — the protégé effect: learning by teaching

The intuition that teaching is good for learning is old, but there is empirical research here too. Educational psychology calls it the protégé effect.

It holds that someone who prepares to teach others, or actually teaches them, understands better and remembers longer than someone who merely studied for the same amount of time.

What is interesting is that even the mere expectation of "I will prepare to teach" produces the effect. The difference begins from the mindset stage, before any actual teaching takes place.

In Nestojko and colleagues' 2014 study, the same material was given to two groups; one was told they would be tested later, while the other was told they would have to teach others later.

In reality both groups only took a test, yet the group that believed they would teach remembered more and organized the key points better. The mindset of teaching changed the very way they processed the information.

Why? Because to teach, you have to restructure information rather than simply memorize it.

You must build a hierarchy of what is core and what is peripheral. You must anticipate where the listener will get stuck. You also have to prepare analogies and examples.

This process is deep processing of information, and the more deeply processed the information, the longer it lasts. The levels-of-processing effect mentioned earlier operates most dramatically in teaching.

I felt this in my body when I was assigned to mentor a new hire at LINE. As I prepared to explain our service's authentication structure to a junior, I stood at the whiteboard and found I could not explain the parts I usually "sort of knew how they worked."

On details like token expiry handling and blocking concurrent logins, my understanding was hollow. Since I normally never had to step on those empty spots, I did not even know they were empty.

That one week I dug back in to teach was the moment I truly understood that system. I had worked beside that code for a year, yet my actual understanding happened during the single week I spent preparing to teach.

One question my junior threw out at the whiteboard that day stays with me. "So what happens if a request comes in at the exact moment the token expires?" I did not know the answer. Thanks to that question, I looked into that boundary case for the first time.

A good student makes a good teacher. The questions you get while teaching pinpoint, with precision, the gaps you would never have thought of on your own.

When you have no one to teach

Many people say, "But I have no junior to teach." Yet the partner does not have to be a person. The key is "pulling it out toward an audience," not whether the audience really exists.

• Empty chair / rubber duck: Explain out loud to a doll on your desk or to an empty chair. It may sound silly, but the moment you make sound, retrieval similar to writing happens.
• Explain to an AI and get rebutted: Ask an AI tool, "I will explain, so point out where I am wrong." The places where you stall while explaining, and the places where the AI questions you back, are your gaps.
• Blog or talk: An in-house lunch session, a short talk, a blog post. As long as there is even one audience member, the density of your preparation changes.
• Record yourself: Record a five-minute explanation into your phone. When you listen back later, the places you glossed over ring out clearly in your ears.

How to teach well — the skill of imagining your audience

I understand that teaching is good, but rambling on any old way does not produce the effect. The key to raising the learning impact of teaching converges on one thing: how concretely you can imagine the inside of your listener's head.

I learned this by accident while studying English and Japanese. Explaining Japanese grammar to a Korean friend and explaining Korean to a Japanese colleague were completely different.

When I explained Japanese particles to a Korean, I could just compare them with Korean particles. "It is similar to eun/neun but subtly different" worked. But when I explained Korean to a Japanese colleague, that comparison did not work, because their starting point was different.

That is when I realized: the same content has to be rebuilt entirely depending on who the audience is. And that "rebuilding" is exactly what deepens my own understanding.

Here are a few skills for good explanation, organized.

• Ask the other person's starting point first: The single question "how much do you already know?" decides half of the explanation. Where I should begin gets settled right there.
• Run it through with one analogy: A good explanation does not switch its analogy halfway. If you likened a distributed lock to a restroom key, finish the explanation inside that restroom to the very end.
• Start with "why it is hard": Do not toss the answer out right away; first show why the naive approach fails. You have to feel the problem for the answer to land.
• Dig in advance where they will stall: While preparing the explanation, mark the spots where you think, "they will surely get confused here." That very prediction is deep processing.

I am reminded of something a coach said when I was learning table tennis. I was struggling with a certain technique, so the coach had me explain it to another member.

As I explained, what was wrong with my own posture flowed out of my own mouth. The coach smiled and said, "When you teach others, you start to see your own form." Even in sports, teaching was a mirror.

The essence of the skill of imagining your audience is, in the end, empathy. To recall what the other person does not know, you have to return to the moment you yourself first did not know it. That round trip ties your knowledge together once more, more firmly.

Overcoming the curse of knowledge

The biggest enemy that makes teaching hard is something else entirely: the curse of knowledge.

Once you know something, it becomes astonishingly hard to recall the inside of your own head from the time you did not know it. It has become so obvious that you forget the very fact that the other person might not know it.

This is why an expert's explanation is often unkind. Not because they are bad people, but because they have forgotten the bridge they themselves crossed.

I too once carelessly said to a new hire, "You obviously know this, right?" Only when I saw my junior's face stiffen did I realize how violent that remark was. What was obvious to me was not obvious to him.

The way to beat this curse is surprisingly simple: first recall, "what confused me the most when I did not know this?"

The moment you restore that memory, the starting point of your explanation naturally lowers. And the work of restoring it becomes yet another act of retrieval that makes you feel back along the foundations of your own knowledge. Here, too, is one more reason teaching becomes learning.

Combining active learning with spaced learning

So far I have focused on "how you pull it out." Now I need to add "when you pull it out." The effect of retrieval grows explosively when it meets timing.

This is where the spacing effect comes in. It is a well-verified phenomenon that, even for the same total study time, spreading it out over intervals leaves far more behind than cramming it all at once.

This is why cramming looks effective right after the exam but vanishes entirely a few days later. Knowledge packed in all at once leaks out all at once.

The spacing effect and retrieval practice are two different principles, but they are most powerful when used together: leave intervals, and at each interval, pull the material out.

The key is "pulling it out around the time forgetting begins." Pull too soon and it is still vivid, so the effect is small; pull too late and you have forgotten completely, so you start from scratch. Pulling it out when it is slightly hazy is what stays best.

The cognitive psychologist Robert Bjork bundles this under the concept of "desirable difficulty." The idea is that a little forgetting and a little struggle actually help learning.

The method I use to memorize Japanese words is exactly this combination. I pull a new word out once the day I learn it, again the next day, again three days after, and again a week after.

What matters is that each time it is "pulling out," not "looking at." Before flipping the word card, I must first recall the answer. Then I check. If the order is reversed, it becomes mere rereading.

Put into practice, it becomes a simple rule. What you learned today, pull out once today, once tomorrow, once this week, once this month. And every one of those pull-outs must be retrieval, not looking.

There is one thing I want to add. Spaced learning is a strategy kind even to lazy people.

You only need a few minutes a day to pull things out, so the burden is small. There is no need to clear a whole day the way cramming demands. Short, frequent, recall. Keep just these three and you are set.

This is how I run my Japanese vocabulary on the morning commuter shuttle. Five words per stop, recalling the meaning before I look at it. If I stall, I mark it and start from that word again the next day. It is not a grand ritual, just a habit tucked into the gaps.

I tried changing my table tennis practice the same way. Instead of hitting one motion a hundred times in one go, I split it across several days. And before each hit, I recall the key point the coach said once in my head before swinging. It is retrieval done with the body.

The active-learning research is real, but those numbers are fake

Here we have to strike a balance. The claim that active learning is good is correct, but among the materials cited to support it, many have dubious sources.

When a good conclusion stands on bad evidence, the conclusion itself ends up doubted. That is why it matters to sift the evidence honestly.

The most notorious is the "learning pyramid," or Cone of Learning. It is a tidy chart of numbers claiming that reading is remembered 10 percent, listening 20 percent, and teaching 90 percent.

It is a regular guest on lecture slides, but the source of these exact percentages has never been confirmed.

It is commonly attributed to the U.S. NTL (National Training Laboratories), yet no one has presented the original data. Educational scholars have long classified this as a baseless "education myth."

The fact that the numbers are suspiciously clean (exactly 10, 20, 90) is itself a warning sign. Real measurements rarely land this neatly. Real-world data is always messy.

The "70-20-10 model" is similar. It is a talent-development framework holding that 70 percent of learning comes from experience, 20 percent from others, and 10 percent from formal education.

This is only a useful heuristic frame in practice, not a precisely measured law. Since it originally rested on 1980s survey recall, taking it as exact ratios is a problem.

Use it as a frame for inspiration, but do not wield it as a scientific law — that is my position. The message "we learn a lot from experience" is right, but "exactly 70 percent" is not.

So what should we believe? Not the tidy pyramid, but effects repeatedly verified through controlled experiments.

The retrieval practice and testing effect mentioned above are such effects. So is the spacing effect.

There is also STEM education research showing active learning beats lecturing. The meta-analysis Freeman and colleagues published in PNAS in 2014 reported that active learning raises exam scores and lowers failure rates.

Keep the conclusion (active learning is effective), but discard the numbers of unknown origin. That is the honest stance. The better the message, the more it must be built on firmer evidence to last.

Comparison — passive vs. active learning

The key row in this table is "immediate feeling." Active learning feels more uncomfortable and inefficient in the moment.

Cognitive psychology calls this "desirable difficulty." Comfortable study usually does not stay; appropriately taxing study does.

The "sense of pace" row is a trap too. Passive learning gives the feeling of advancing quickly, but since little remains, you end up having to review again. Active learning looks slow, but because it lodges deep in one pass, it actually comes out ahead in total time.

Common excuses vs. rebuttals

There is always a plausible excuse for not switching to active learning. Here are the excuses I have heard from myself and from juniors, along with rebuttals.

Making this table, I realized something. Most excuses are pretexts for avoiding the "discomfort" that active learning brings. And that discomfort is precisely the signal of its effect.

Practice — a routine that turns 70 percent of input into output

No grand overhaul is needed. You just shift part of your existing input time into output.

1. Force an output ratio into your input: After reading 30 minutes, close the book and spend 10 writing a summary. Input 45 plus output 15 leaves more behind than input 60 alone.
2. Blank-page retrieval at the end of each chapter: Whenever a chapter or lecture ends, close the book and write three key points. If you cannot, reopen only that part.
3. Schedule a weekly "teach": Each weekend, pick one thing you learned that week and explain it for five minutes to a partner — a person, an AI, or an empty chair.
4. Leave one piece of writing per month: Once a month, organize one thing you learned into a short piece. The internal wiki or a blog both work. Assume there is a reader.
5. Retrieve with spacing: Retrieve the same content again a day later, a week later, a month later. Combining spaced learning with retrieval is the most powerful.

An example of input vs. output time allocation

The same 60 minutes leaves different results depending on how you use it. Below is an example of the allocation I recommend.

The key is the ratio. The point is not to make input zero. Just changing the ratio from 100 to 0 into 70 to 30 or 60 to 40 already changes the result markedly.

An example weekly routine

The charm of this routine lies in how light the burden is. Three lines every night, five minutes on the weekend, one piece at month-end. Even added up, it is not much time. Yet these small outputs turn ordinary input into knowledge that stays.

Pitfalls — what happens when you misread active learning

• Discarding input for output only: You must not ignore input just because active learning is good. You need something to pull out before you can retrieve. Input is still necessary; it is only a question of ratio.
• Mistaking highlighting for active learning: Underlining, transcribing, and pretty note-taking look active but are usually passive. The hand moves, yet no memory is pulled out. The test is "did you pull it out?"
• Mistaking discomfort for inefficiency: Active learning feels slow and taxing, so it is easy to flee back to rereading, thinking "this is inefficient." That discomfort is often the very proof of the effect.
• Being swayed by numbers: Do not base your case on remarks like "teaching means 90 percent retention." Baseless numbers may motivate, but they are not the truth. Standing on honest evidence is more advantageous in the long run.
• Failing to start because you want perfect writing: Learning writing is not for presentation. It can be bad. The point is pulling it out of your head, not writing it well.
• Retrieving too soon: If you read something and pull it out immediately, it just gets copied from short-term memory. Pulling it out after leaving a little time to forget is stronger.

FAQ

Q. So is reading useless? No. Without input there is no output. Reading is the stage of gathering materials, and writing and teaching are the stage of making those materials yours. Both are needed; this is just a story about increasing the output share.

Q. I have no time, and you want me to output too? Output is not extra time but a reallocation of input time. Instead of reading the full 60 minutes, read 45 and organize for 15. The total time is the same, but what remains differs.

Q. I really have no one to teach. A person is not required. Explaining to an empty chair, a rubber duck, or an AI still triggers the retrieval effect. The key is "pulling it out by voice or by hand."

Q. Can I use the learning pyramid numbers in a talk? I would not recommend it. They are statistics of unclear origin. Cite verified evidence instead, like retrieval practice or Freeman's active-learning meta-analysis. It is more persuasive and more honest.

Q. I did a blank-page summary but could barely write anything. Is my method wrong? On the contrary, your method is right. The blank you cannot fill is exactly where you did not know. You found a gap that rereading would have hidden forever, so reread only that part and pull it out again.

Q. Does this work for studying a foreign language? In my experience it is one of the fields where it works best. Recall a word before you look at it, and try explaining grammar to someone else. I saw the greatest effect studying Japanese this way.

Q. Does it apply to sports or hands-on skills? The form differs, but the principle holds. Try explaining your posture to someone, or writing down the key point of today's motion. While learning table tennis, I often discovered my own problems while explaining my form to others.

Q. Is it not faster to have the AI explain it to me? Why must I be the one to pull it out? Hearing the AI's explanation is, in the end, just another input. It is fast and smooth, which is exactly why it does not stay. The AI is most powerful not as a tool that gives answers but as a tool that grades the answer you pulled out first. Reverse the order. Explain it yourself first, then ask the AI where you went wrong.

Closing — let us do study that stays

Ever since the experience of that distributed-systems book scattering within a month of finishing it, I changed how I study.

I reduced how much I read and increased how much I pull out. I close a chapter and write on a blank page, explain to a junior, and organize it on the blog.

My pace slowed. If I used to get through one book a month, now I do not even finish half a book. But far more remains. Half a book lodged in my head beats a whole book scattered.

At first this slow pace made me anxious. Everyone else seemed to be racing far ahead while I alone was marking time.

But as time passed, I came to understand. A book read fast and forgotten was not much different from a book never read. The half I read slowly and kept was a real asset I could still answer a colleague's question about a year later.

Progress is visible; forgetting is invisible. That is why we keep getting fooled by fast progress. Once you factor in the invisible forgetting, slow active learning is actually the faster road.

The biggest thing I learned is this: be suspicious of study that feels comfortable.

That feeling of reading smoothly and nodding along is often familiarity, not understanding. Real study lives in that uncomfortable place where you stall at a blank page and stumble while explaining.

And the best tools for enduring that discomfort are writing and teaching. Both are, in the end, acts of "pulling out what I know."

Knowledge you have not pulled out is not yours. The mere feeling that it is in your head is not enough. Only after you pull it out can you find out whether it is really yours.

If you read something today, close the book and write even one paragraph in your own words. That one paragraph will stay longer than the chapter you simply read and moved past.

It was the same with this post. While pulling out and organizing what I had read about active learning into a single piece like this, I felt the scattered concepts finally bind into one inside me. Stop reading, and now you too, please pull out one line.

References

• Karpicke, J. D., & Roediger, H. L. (2008). The Critical Importance of Retrieval for Learning. Science. https://pubmed.ncbi.nlm.nih.gov/18276894/
• Freeman, S. et al. (2014). Active learning increases student performance in science, engineering, and mathematics. PNAS. https://www.pnas.org/doi/10.1073/pnas.1319030111
• Nestojko, J. F. et al. (2014). Expecting to teach enhances learning and organization of knowledge. Memory & Cognition. https://pubmed.ncbi.nlm.nih.gov/24845756/
• Brown, P. C., Roediger, H. L., & McDaniel, M. A. (2014). Make It Stick: The Science of Successful Learning. Harvard University Press.
• Bjork, R. A. — Desirable Difficulties in Theory and Practice. (Source of the desirable-difficulty concept)
• Critique of the Learning Pyramid myth — Will Thalheimer analysis. https://www.worklearning.com/2006/05/01/people_remember/
• The Feynman Technique explained — Farnam Street. https://fs.blog/feynman-technique/
• James Clear — The Spacing Effect and learning. https://jamesclear.com/spacing-effect
• A critical review of the 70-20-10 model — HBR. https://hbr.org/2016/07/where-companies-go-wrong-with-learning-and-development