Big Idea

The argument in one line.

AI teaches you fastest not when it explains things, but when you use it to interrogate, map, quiz, and spar with yourself, because learning is a production skill, not a consumption one.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You use AI mostly to have things explained to you and want other ways to actually retain what you learn.
You're trying to get fast at a specific skill (an interview, a new tool, a technical subject) and want a repeatable process.
You like structured frameworks and prompts you can copy directly into a chat window.

SKIP IF…

You're looking for a specific AI tool or app recommendation rather than a general learning method.
You already run a rigorous study system (spaced repetition app, tutor, structured course) and have no interest in AI chat tools.

TL;DR

The full version, fast.

Most people only use AI as an explainer, the most passive of ten roles it can play in learning. Real learning comes from production: actively recalling, predicting, and explaining, not rereading or being told an answer. Reread a document four times and you remember 83% after five minutes but only 40% after a week; read once and test yourself three times and you remember 71% then 61%. The other nine roles push toward production: the interviewer clarifies your real goal, the mapmaker builds a curriculum, the Socratic questioner and examiner expose gaps, the checker grades your process, the diagnostician finds your recurring root mistake, and the sparring partner drills real reps under pressure.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 03:42

01 · AI Learning Principles

Human memory is built on active production (recall, predict, explain, attempt), not passive consumption (reading, watching, being told). Retention data: rereading 4x gets 83% recall at 5 minutes but 40% at a week; reading once plus testing 3x gets 71% then 61%.

03:42 – 06:04

02 · AI Learning Roles

Most people only use the explainer role. Nine other roles exist, nearly all oriented around production rather than consumption: interviewer, mapmaker, Socratic questioner, examiner, checker, listener, diagnostician, sparring partner, clerk.

06:04 – 07:54

03 · Spaced Repetition Strategy

The Ebbinghaus forgetting curve shows memory decaying after learning; spaced repetition (periodic active recall) flattens that decay and raises the retained baseline over time. AI can generate flashcards, schedule reviews, and convert mistakes into new cards.

07:54 – 11:23

04 · Interviewer and Map Maker

The interviewer asks what the goal is, current knowledge level, and how you'll be tested before teaching anything. The mapmaker breaks a shapeless subject into sub-topics, dependencies, and common sticking points.

11:23 – 13:26

05 · Explainer and Review

The explainer illuminates one stuck step, but the learner must then redo the whole task from scratch without help, the redo is what converts explanation into retained knowledge, at a cost of only 10-15% more time.

13:26 – 14:36

06 · Socratic Learning Loop

The Socratic questioner never hands over the answer, instead asking follow-up questions that force the learner to surface their own gaps, looping back to mapping and explaining as new gaps appear.

14:36 – 17:01

07 · Examiners and Checks

The examiner quizzes at escalating difficulty until the learner starts failing, marking the edge of their knowledge. The checker grades the process behind a piece of work (a summary, proof, or code), not just the final output.

17:01 – 19:56

08 · Listener and Diagnosis

The listener grades a learner's spoken, written, and drawn explanations against the source material. The diagnostician reviews many past conversations to find one recurring root misunderstanding behind otherwise unrelated mistakes.

19:56 – 22:08

09 · Sparring for Practice

AI role-plays interviews, sales calls, and negotiations with a timer and an adjustable difficulty slider, giving the learner real reps under pressure before the real event.

22:08 – 23:05

10 · Clerk and Organization

The clerk turns messy notes or intermediate work into a clean outline or flashcards without adding new content, multiplying the value of spaced repetition for the same time invested.

23:05 – 25:33

11 · Learning at the Right Level

Have AI quiz from easy to hard and stop at the point you start guessing to find your real current level, explain a concept at three levels to find what clicks, and pull a short list of resources matched to that level.

25:33 – 27:11

12 · Simulations and Practice Apps

AI can simulate realistic tasks (interviews, customer emails, negotiations) and grade the deliverables produced, and coding agents can cheaply build small practice apps and flashcard tools for extra repetition.

27:11 – 29:36

13 · When AI Hinders Learning

Letting AI explain something feels like understanding but isn't, similar to how GPS use correlates with weaker spatial memory. Watch for AI agreeing when you're half right, inventing sources, or handing over finished answers; the fix is using AI to support the struggle, not skip it.

Atomic Insights

Lines worth screenshotting.

Learning is a production activity (recalling, predicting, explaining, attempting), not a passive consumption activity (reading, watching, being told).
Rereading a document four times gets 83% recall after 5 minutes but only 40% after a week, while reading once and testing three times gets 71% after 5 minutes and 61% after a week.
Most people use AI only as an explainer, which is the single most passive of ten distinct roles AI can play in learning.
The interviewer role has AI ask what you actually need before teaching anything, turning a vague goal like 'learn coding' into the real goal of 'pass this specific interview in three weeks.'
The mapmaker role turns a shapeless subject into a concrete curriculum: core sub-topics, what depends on what, and where people typically get stuck.
Skipping the redo-it-yourself step after AI explains something means you've only read about the solution, not learned it.
The Socratic questioner withholds the answer and asks questions instead, exposing gaps in understanding the learner didn't know they had.
The examiner scores a topic at escalating difficulty until the learner starts failing, which pinpoints exactly where their knowledge ends.
The checker grades the process behind an answer, not just the final answer, the same way a math teacher requires showing your work.
The listener compares a learner's spoken, written, and drawn explanations of a topic against the source material to catch gaps a single format would hide.
The diagnostician reviews dozens of past AI conversations to find the one root misunderstanding causing unrelated mistakes across different subjects.
The sparring partner role-plays high-stakes situations like interviews and sales calls with an adjustable difficulty slider, forcing real reps under time pressure.
The clerk turns messy notes into clean outlines or flashcards, which can multiply the value of spaced repetition 5-10x for the same study time.
The fastest skill growth happens in the zone of proximal development, material just outside what a learner can already do alone.
People who rely on GPS develop weaker spatial memory, and the same shortcut effect applies to reasoning when AI is used to skip the struggle instead of support it.
AI that agrees when you're half right, invents sources, or hands over finished answers turns learning into collecting notes instead of building understanding.

Takeaway

AI has ten learning roles, most people only ever use one.

AI LEARNING ROLES

Real learning happens through active production, not passive explanation, and AI can support production through nine roles beyond the explainer that most people never try.

01AI Learning Principles

Learning happens through production (recalling, predicting, explaining, attempting), not through consumption (reading, watching, being told).
Rereading material four times gives 83% recall after 5 minutes but only 40% after a week; testing yourself after one read gives 71% then 61%, a much smaller one-week decay.
The gap that matters is long-term retention, not how much you remember five minutes after studying.

02AI Learning Roles

Most people use AI only as an explainer, the single most passive of ten distinct roles it can play in learning.
The other nine roles (interviewer, mapmaker, Socratic questioner, examiner, checker, listener, diagnostician, sparring partner, clerk) are built around production, not consumption.

03Spaced Repetition Strategy

Memory decays on an exponential curve (the Ebbinghaus forgetting curve) unless you review it; each spaced review flattens that decay and raises the retained baseline.
AI can generate flashcards, schedule reviews, and convert mistakes from a sparring or listening session into new cards automatically.

04Interviewer and Map Maker

Before asking AI to teach you something, have it interview you: what's this for, what do you already know, and how will you be tested.
Vague goals produce vague study plans; specify the real constraint (interview in 3 weeks, on LeetCode, 5 minutes per question) so AI skips what you don't need.
When a subject feels shapeless, ask AI to map the sub-topics, their dependencies, and where people typically get stuck before you start studying.

05Explainer and Review

After AI explains a stuck step, redo the entire task yourself from the start without help, or you've only read about the solution, not learned it.
The redo pass costs only 10-15% more time than the first pass but is what actually converts explanation into retained knowledge.

06Socratic Learning Loop

A Socratic questioner withholds the answer and instead asks follow-up questions, which surfaces gaps in understanding you didn't know you had.
Loop between mapping, explaining, and Socratic questioning as new gaps appear, rather than treating any single pass as finished.

07Examiners and Checks

An examiner quizzes you at escalating difficulty until you start failing, which marks exactly where your knowledge ends.
A checker grades the process behind your work, not just the final answer, the same way a math teacher requires you to show your work.
Checking your process, not just your output, can reveal a shorter or faster path to the same result, speeding up every future attempt.

08Listener and Diagnosis

Explain a topic back in three formats, spoken, written, and drawn, and have AI grade all three against the source material; each format exposes different gaps.
A diagnostician reviews your last several dozen conversations to find the one recurring root misunderstanding causing seemingly unrelated mistakes across different subjects.

09Sparring for Practice

Use AI as a sparring partner for interviews, sales calls, or negotiations: give it a scenario, a role, and a time limit per answer.
Turn up the difficulty slider as you improve, forcing a higher standard than you'd get from an easy conversational partner.

10Clerk and Organization

Use AI as a clerk to turn messy notes into a clean outline or flashcards without adding new content, multiplying the value of spaced repetition 5-10x for the same study time.

11Learning at the Right Level

Have AI quiz you from easy to hard and stop at the point you start guessing, that's your actual current level, not the level you assume.
Ask AI to explain a concept at three levels (child, beginner, expert) to find the explanation that matches where you really are.
The fastest growth happens in the zone of proximal development, material just outside what you can already do alone, not material that's too easy or too hard.

12Simulations and Practice Apps

Role-play realistic scenarios (interviews, customer emails, negotiations) and have AI score the deliverables you produce, not just your talking points.
AI agents can build small practice apps and flashcard tools cheaply, giving you another modality to produce and test knowledge.

13When AI Hinders Learning

Letting AI explain something feels like understanding but isn't, the same way GPS use correlates with weaker spatial memory.
Watch for AI agreeing when you're half right, inventing numbers or sources, or handing you a finished answer; each one turns learning into passive note collecting.
The fix isn't avoiding AI, it's using it to support the struggle (production) instead of skipping it.

Glossary

Terms worth knowing.

Zone of proximal development: The band of material just beyond what a learner can already do alone, where guided practice produces the fastest growth.
Ebbinghaus forgetting curve: A model, over a century old, showing memory strength decays roughly exponentially after learning unless the material is reviewed.
Spaced repetition: Reviewing material at increasing intervals timed to just before you'd forget it, which flattens the memory-decay curve over time.

Resources

Things they pointed at.

29:02productMaker School ↗

29:02productMaker Zero (free course resources) ↗

Quotables

Lines you could clip.

00:42

“Learning is almost entirely a production thing. It's not a passive consumption thing.”

states the video's whole thesis in one line→ TikTok hook↗ Tweet quote

28:00

“If it explains something to you, it'll feel like understanding, but it isn't.”

sharp, counterintuitive, one sentence→ IG reel cold open↗ Tweet quote

28:18

“People that use GPS apps tend to have poorer representations of their environment in their brains, and I think AI is quite similar.”

vivid analogy that reframes the whole warning section→ newsletter pull-quote↗ Tweet quote

03:26

“Rereading it a bunch of times, in a week they only remember 40%, whereas actively producing and testing gets you closer to 61%.”

concrete stat that backs the thesis→ TikTok hook↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

How is it that some people can learn in just a day what other people take weeks or months to learn? And how can we use AI to positively augment our learning as opposed to confuse us, reduce our focus, and destroy our attention spans? Well, in this video, I'm going to show you everything that you need to know about how to use AI to learn anything in the modern era in just a fraction of the time.

I use these tactics in order to build a business that does over $400 ,000 a month. I also went to school for behavioral neuroscience, and this is exactly what we focused on. It's how to use technology to augment and improve our ability to learn.

It was by far the subject I was most interested in. And so I use this sort of stuff all the time to help people, yourself included, learn things faster. So before we get into how to actually use AI to learn, it's important to understand how human beings actually learn.

Because I feel like a lot of people... you know, they misunderstand this part. And if you misunderstand this part, it will lead to a bunch of downstream issues and, you know, productive and effective learning.

You know, most people believe that you learn through consumption, which is where you will read something, right? You will watch something like this video, or you'll get it explained to you, say via an AI bot or a chat bot. And the issue is this feels like learning, right?

But the way the human beings actually learn, the way the, you know, memory encoding mechanisms in our brain and the retrieval mechanisms in our mind actually work is based off of production. And so this involves recalling things actively, predicting things actively, explaining things actively, and even attempting to do things out there in the real world.

Okay. And this feels way harder, which is why our brains constantly trying to preserve energy, avoid that. But when you understand this distinction, you also understand what leads people to being extremely effective learners versus ineffective learners.

You know, there've been a bunch of studies done on this subject. One of them showed that, you know, if you were to reread some source document four times, and then get quizzed on it five minutes later, you would remember 83 % of the source document.

So you'd think, wow, I've only lost 17 % of that. That's pretty good, right? Likewise, if you read a document once and then you are tested on that three times immediately after, five minutes later, you'll remember 71%.

But the thing that we're most interested in is not how much are you going to remember in five minutes, right? The thing that you're most interested in is how much are you going to remember in a week or a month or a year? How deeply will the things that you've learned stick with you for the rest of your life?

And when a person rereads some source document four times, which many of us will do, right, thinking that rereading the source document will teach us things and make us learn it, in a week they only remember 40%, whereas actively producing and testing gets you closer to 61%, meaning there's a decay of only 10 % in the read once, test three times group, whereas there's a decay of, I mean, geez, at least 50 % in the rereading it a bunch of times group.

Okay, so I'm sure people have probably talked all day about this sort of result to you, and it's just glossed in one ear, out the other. But the reality is active production is ultimately how you learn. And so the whole idea of using AI to learn faster and better is actually augmenting your ability to actively produce things, not just necessarily passively read things.

There are many different roles that artificial intelligence can play for us. The issue is most people will only have ever used one. They use the explainer role.

which is where you've tried to understand something, you're a little bit stuck. And so maybe you paste in the source document to AI and you say, hey, I'm stuck. Can you help me out?

Explain this in a different way. The issue with that is, first, as we see already, that's a passive consumption mechanism. You're just reading.

The second is that that's only one of about 10 different roles that an AI can help you play. And those other nine roles are primarily focused on your ability to produce information. So if all you've ever done with AI is you've had to explain things to you in a really simple manner, not only are you only scratching the surface of what is truly possible in consumption versus production, but you also have nine other fundamentally different mechanisms that you're just leaving on the table.

And so I'm going to cover these nine mechanisms. The first is the interviewer. That's when you're starting out and you're unsure of what you need.

You don't even know enough to know what you don't know. So you just have an AI who has a shallow knowledge of virtually everything in existence, ask you questions to determine what direction you want to go, what your state of learning is right now, and so on and so forth. The mapmaker is where, you know, you're just starting a subject.

It feels really shapeless. You don't have a defined curriculum in place. And as we know, if you put a roadmap in front of most people, they'll be able to do a thing.

But if they don't have a roadmap, they don't know where to go. Oftentimes they're lost. They'll just be swimming, you know, listlessly in the water.

Well, the mapmaker can help you figure out where to go. It can essentially produce a curriculum for you. I'll show you guys how to do that in a moment.

How about the Socratic Questioner? This is where you think you understand a topic. The Socratic Questioner allows you to better develop your understanding of the topic by forcing you to actively produce.

Or an Examiner. You know, you've covered a topic fully. You want to figure out how much you actually get of a topic.

You want to score yourself with some percentage. Well, that's what AI can do for you. It can literally be essentially an oral or verbal or...

written examiner, just like you used to have back in the good old days, you know, 16, 1700s university where you'd be doing your thesis defense in front of a bunch of people. You can do that now with AI. How about a checker?

You know, maybe as part of your learning, you've made something, but you can actually have AI verify certain domains to ensure that your knowledge was effective in producing the outcome that you want. Maybe a listener where you explain it back, but in your own words, AI then helps you determine the gap based off of its understanding of the topic more generally.

There's diagnosticians where, you know, if you have the same mistake that keeps on coming up in your learning, you can now actually ask your AI agent and I'll show you how. Hey, you know, I want you to go through the last five or 10 or 20 conversations we had about the subject. What am I persistently getting wrong?

What do I persistently need to understand to patch these holes in my knowledge? How about a sparring partner? A lot of the time you're learning a skill because it is a real world applicable thing.

You want to public speak. Maybe you're preparing for an interview. Maybe you're trying to crush the leak code circuit.

Well, you can actually have a sparring partner with which to assist you on that and basically simulate things. And finally, clerk, how many hours have you spent just coming up with notes and trying to keep track of all this information that you've digitized? You don't need to do that anymore.

AI can do all of that for you. And so what I'm going to do is I'm going to show you guys how to apply AI in all of these different ways and ultimately improve your learning from a scientific perspective, not just a thing that sounds nice like most other people will. A really simple example of this is something termed spaced repetition.

For those of you guys that don't know, if the y -axis here is how strong your memory of a subject is, and then the x -axis is the amount of time since you learned the thing, you can see that when you learn the thing initially, you have a lot of memory for that thing. You've really consolidated that thing. But when you don't review it periodically, what happens is the strength of that memory decays, almost as like this inverse exponential.

And so spaced repetition... is the process whereby you consult frequently with the thing that you learned. You actively produce that knowledge.

Let's say you're trying to learn how a car engine works. And then based off of what's called the Ebbinghaus forgetting curve, which is this psychological idea developed 100 and something years ago, you test yourself on a... structure on a basis every now and then using, you know, AI or another tool to ask yourself, how the hell does this engine actually work?

Okay, the pistons, the cylinders, what are the four stages of combustion, you know, intake and so on and so forth. And then in that way, your memory actually gets extremely strong. And as you see, every time you, you know, repeat this spaced repetition thing, which I'm going to talk a little bit about in a moment, two things occur.

One, the memory lasts a lot longer. You can see the slope of this curve is actually a lot less steep. And two, it never dips as far down.

And so the strength of the base memory just gets stronger and stronger. This also means that in just a few moments of studying a topic, you can usually remember it far more than if you were to like wait this whole time and have to restudy it again from scratch, right? So that's just one of many ways that you can use AI.

AI can write your flashcards for you. It can keep a schedule or you can use an app like Anki. It can turn any mistakes you make during a sparring session or some sort of listening session into new cards for you to review.

I'm going to give you guys some prompts and stuff like that that you can use for that sort of thing as well. But the whole idea is this is one simple example of how AI can help you. So if you don't know enough about a topic to even begin knowing what you know, what you don't know, let's chat about the interviewer.

The interviewer is the simple idea where you have a chat bot up here on the left, you have you on the right, and basically you're saying, hey, I really want to learn X, Y, and Z thing. And instead of just saying, help me learn X, Y, Z thing, you say, could you drill me on anything and everything under the sun to determine how much I already know about X, Y, Z thing?

And then in conjunction with the map maker in a second, what I need to learn in order to patch the holes in my knowledge. And this makes sense that AI would be good at this, right? I mean, it's a massive, compressed form of basically the whole internet.

That is essentially the whole point of AI, really, just to point you in the right direction, to act as a fast way of approximating knowledge. And so, you know, if you're like, hey, I want to learn, you know, coding, AI might ask you, okay, what is this for? Well, it's for a job interview in three weeks.

Now the AI knows, okay, it's not actually that you need to learn all of coding. You need to learn enough coding to make a job interview work. Okay, so what do you know already?

Well, I know some basic Python. You know, maybe I spent two weeks doing some bootcamp a while ago. Okay, how will you be tested?

Well, live coding with some sort of timer. From here, AI is determining a variety of different things to help you accomplish your stated goal, not just the goal that you might think you have. Okay, and so in this case, the goal is not actually learn coding.

The goal is pass the interview. You don't need to learn Python basics because that's a waste of time. You've already gotten to that point.

In order to practice and evaluate and assess your learning, A simplest and easiest way to do that is going to be some sort of timed live coding because it's going to mirror the actual examination conditions. This is all stuff you would have figured out, you know, eventually if you were a perfectly rational agent, but it's stuff that you probably wouldn't know enough to know right up front.

So if you are starting out and unsure of what you need, you can just ask the AI, hey, you know, before you teach me anything about coding, could you ask me questions about my goal, my level, everything that I need to know basically in order to make this work really effectively? And the AI model will go, absolutely. A couple of things to watch for here, vague answers, vague plans, you know, ultimately want to be specific.

So it's not about, oh, you know, my goal is to pass an interview. It's like, oh, my goal is to pass an interview that's going to be taking place on leak code. You know, I know a little bit of basic Python, you know, it's been two weeks getting all the way up to, I don't know, like HTTP requests or something like that.

The timer is five minutes per question, and we're going to be pulling from this question bank, right? Like you want to be more specific than just that. And that takes me to the idea of a map maker.

So the reason why it's very important to have a map or a plan in place is because if you don't, the vast majority of your time and energy will simply be spent like fighting off the almost emotional anxiety of not knowing what you are going to do next. And so you can eliminate that completely by just giving yourself something to do next, rather than finishing a module and then wondering, okay, what the hell am I going to learn next?

What fits? You know, just build a plan and then you can follow a plan. Even if it's not perfect, you're going to be a lot more productive as a result.

So let's say you want to learn statistics. That's a pretty damn broad term. There's a lot that goes on in statistics, right?

So if you just say, I want to learn statistics, that's pretty shapeless. What AI can help you do is it can help you break that down into the core fundamental sub -modules of statistics. And so maybe you'll start with averages.

Then you'll learn about spread and probability. Maybe after that, you'll learn about distributions, sampling. And then finally, loop around to some sort of A -B test.

This is what AI can essentially allow you to do. If a subject feels really shapeless, what I will always do before I even try learning is, hey, just map out statistics for me. I want the main parts.

I want the various things that depend on a bunch of other things. And I want to know where people usually get stuck on so I can plan my time ahead of time. And that takes me to the explainer.

Many people here are already using the explainer day to day. You know, they'll have read how to do something. They'll even have set up how to do something.

When it comes time to actually solving the problem that they initially wanted to solve, they'll be stuck in it. The explainer just helps you illuminate just that one part for you, explains, you know, how the solution works or something that allows you to then, you know, learn it faster. The issue is most people don't do this next step, which is redoing the initial thing you wanted to do without help.

Remember that learning is almost entirely a production thing. It's not a passive consumption thing. If you don't know something and you ask, hey, can you help me figure this thing out?

And then it explains it to you and you nod along and say, yeah, that makes sense. Okay, cool. I've now learned the thing without actually assessing.

from a baseline, whether or not you know the thing without the AS help, you have not learned that thing. You have merely read about that thing. And there's a very big difference between the two, right?

So after you've gone through a whole process once, it is integral that you revert back to the very first step, which in your case might be reading about how to set this thing up, and actually walk through it from first principles from the start in order to go from zero all the way up to one. That's the way you demonstrate your knowledge, and that's the way that you actually productively remember it.

And so anytime I'm trying to learn, let's say a new AI tool or something like that, I will use AI quite heavily to interview me on my current state of knowledge to create a map for me on the various concepts I need to know in order to learn it quickly. And then I'll actually try. And if I struggle, I'll have AI explain things to me.

But I don't just stop there. I'll actually... Once I've figured it out and we've made it all the way to the end of whatever task I have, revert all the way back to the beginning and then try doing it all myself.

This only will ever take you a few additional moments of time, 10 or 15 % of the total length of time you spent studying a subject. But the end result and your ability to know, learn, and then articulate it to other people like I'm doing to you right now is insane. It's unmatched.

So you need to redo things without help. You can't just, you know, try and one -shot things progressively as you go. You really do need to loop back.

And that's ultimately how to use an explainer. Simple thing to say here is, hey, I tried doing XYZ. I got stuck at XYZ.

Could you explain just that part at my knowledge level so that I understand it? Once you're done with that, again, circle back and then finish the rest. The Socratic questioner is also extremely powerful.

You know, you, a lot of the time, are going to think that you understand stuff. Again, because AI will have explained something to you. You'll have gone through it at least once.

And you'll say, yeah, I think I get the gist of it. But the Socratic Questioner is a way for AI to basically act as a study buddy with you and ask you things like, okay, so why does this statistical principle work? Well, the reality is, if you don't know how to answer it, it will immediately have clarified to you where your gaps in knowledge are.

And then if you don't know, maybe it can ask you a follow -up question like, okay, what if the input to the statistical thing is doubled? Then you'll say, okay, then that would break. Now the AI is learning the gaps in your knowledge.

You're also seeing the gaps in your knowledge. And then that can help direct you back to the explainer. to learn exactly how that process works.

And then you can circle back and forth between Socratic questioning, explanations, creating a new map to learn new things that you didn't realize you didn't know, and so on. This allows you to find the gap yourself. You now know the answer.

It'll never just hand it over to you because it's now asking questions productively, forcing you to come up with things. And then as a result, you can actually understand something, not just think you understand. So how to try that is, you know, hey, I want to learn X, Y, and Z thing.

Don't just give me the answer. Ask me questions until I find what I don't know myself. Next up is the examiner.

Now, there are a lot of situations where you think you understand something, okay? And then because you don't test yourself, you don't actually realize where you fall short. But with an exam that is punctuated at varying levels of knowledge, what you can do is you can actually figure out, okay, I figured out level one.

I understand how that works. I know level two. I get level three.

Level four makes sense. Level five, didn't actually pass that test. Why?

And so it's another way of you figuring out where you need to go next in order to maximally direct your learning. And that's all ultimately this whole process is about. It's about maximally directing your learning.

Human beings have so much potential buried inside of them. Our brains are so incredible. But simply because of like old deprecated educational institutions, formats like lectures, which were developed thousands of years ago and just do not scale into our current era, and then other gaps in knowledge and so on and so forth, we're just not capable of squeezing out more than 10 or 20 % of it.

So this helps you do that. It's very straightforward and easy. A simple way to do this is saying, hey, could you quiz me on this topic?

I just want one question at a time. Make every question harder until, you know, I don't get things right. And then at that point, you'll know where the gap in your knowledge is.

How about the checker? In this situation, there are a lot of places where, you know, you don't learn for the sake of learning. A lot of the time you learn for the sake of creating something.

And so what the checker does is basically scores you on almost like a rubric similar to an examiner, but this time checking, you know, whether or not your process works to begin with. And so the value here is you're not assessing just the end result of your knowledge. What you're doing is you're assessing the process required to create the knowledge itself.

And so, you know, I don't know if you guys have ever taken a really introductory math class. A lot of the time they don't just say, don't just give me the answer. I actually want to know the process behind your work.

So show your work. Show me, you know, how you multiply the two variables together or whatever. What the checker does is it assesses the intermediate steps to determine that your logic and your reasoning was correct in order to get there.

In this way, you know, if you have any issues, the model can figure this out. So if you made something like a summary, a proof, some sort of code, a math problem or something like that, actually have it evaluate or check your reasoning to see if there's a shorter and faster way you could have made it to the same point.

In this result, you know, maybe we could have shortened the summary down twice, got it literally to about half the length. And as a result, the next time you encounter this problem, your process itself will be faster, not just the end result. So in this case, you might say, hey, here's my work for determining X, Y, and Z.

Can you find what's wrong or missing and let me know if there's a faster or easier way to get there? Don't rewrite any of it. Next up is this idea of the listener.

So the end result is, you know, you've now learned something to a reasonable amount of time, but you have to understand that there are different ways to demonstrate your knowledge. There's like a written way of doing it. There's like a verbal way of doing it.

I mean, there's like five or 10 and the best learners know how to provide that knowledge in multiple different formats. And so it's not enough just to, you know, say something or write something, you know, with your fingers. My recommendation is if you really want to maximize all the different highways of your mind, you know, you should say something.

You should visualize something with maybe some sort of graphic. You should write something. You should really try and engage yourself maximally across all senses.

What the listener does is it compares your words across all of these to the source document on all of these different testing mechanisms, on the verbal reasoning, on like your ability to write things out and your ability to visualize things and so on and so forth to determine whether or not you were correct. And so what I will typically do is when I really want to make sure I understand a topic, I will just hold down my voice transcription key and I'll just start talking into, you know, my computer.

After I do my voice transcription, I will try and draw a quick image of the concept to make sure I understand. And in addition, I will write a much more. short and succinct version of my explanation since verbal voice transcriptions and stuff like that tend to be a lot longer than their written counterparts.

I'll then compare it to the source document, whatever it is I'm trying to learn. Maybe it's some new neural network architecture or something like that. And now it has three different means by which to assess whether or not I understand something, the visual, the verbal, and then ultimately the written.

And it can compare all three of these to the source document. And as a result, it can listen to me and determine whether or not I've made it. So here maybe you'll say, hey, I want to explain this idea in my own words.

I want you to grade it against the source and tell me exactly what I missed. The diagnostician, you know, despite being kind of a tongue twister, is a very important part of understanding. Essentially, a lot of the mistakes that you make in learning occur at the same fundamental roots.

Maybe you just don't really understand some simple mathematical operation or something. And as a result, you don't understand how to do a specific thing in statistics. You don't understand how to do a specific thing in biology.

And then you don't understand how to do a specific thing in chemistry. Let's say you're just in a university environment trying to learn all this stuff. The reality is you don't actually know what's holding you up, despite the fact that these are all templated issues.

So what the diagnostician does is All of these things are the result of a shared misunderstanding. What you do is you just ask the model, hey, like what is the recurring mistake in reasoning or logic that I keep on making across all of these, you know, listenings, examinations, you know, checks and so on and so forth.

Read through my last, you know, 100 conversations and figure this out. And it'll actually go and it'll find the core reasons why you misunderstand stuff. It will literally go and tell you like, you appear to be making the same sort of logical leap from X to Y.

The reason for that is probably that you don't understand Z. And now you, instead of trying to learn the specific, outcome of each can just learn the root thing that keeps on making you make the same mistake.

So, you know, here are a bunch of things I keep getting wrong. What misunderstanding do all of these things have in common? Just feed that to your AI agent and you will be far better off.

The sparring partner is extremely valuable. And I personally have been using this a lot, especially as I've started doing more podcasts, interviews, and then high stakes negotiations in business. So I will often actually have AI give me some sort of back and forth Q &A.

You know, if I'm like, hey, I have an interview coming up for X, Y, and Z, I want to spar back and forth, you know, verify that I know what I'm doing. I'm going to do this verbally. And then I also am going to set a timer with every question to try and figure it out within 30 seconds, because I think that's probably how much time I'm actually going to be given.

Let's do it. Hey, why should we hire you? Well, it's because I work hard, right?

You send that back to an AI agent that you've pre -primed as a sparring partner, it's probably not going to be too happy. And it's right, right? So let's say something like too vague.

Give me an example. Well, last month I cut our wait times in half. So maybe that's what you should have ultimately led with.

The sort of sparring in process is, it's very difficult to do correctly. And you have the ability to adjust that difficulty slider. You can reduce the amount of time you have to come up with each question.

You can obviously, you know, force the AI model to be a higher and higher and higher level interviewer as you go across. But the result is you actually end up getting a couple of reps on it, but actually having to do the reps. And I mean that in a good way.

You know, a common thing that I'll teach people how to do is I'll teach them how to do, you know, sales calls for AI and automation services, because that is part of what I sell and what I show people how to sell on this channel. And so, you know, they will have never jumped on a sales call before in their entire life. They'll be absolutely terrified.

You know, they don't know what to say. They don't know how to frame the call. They don't know how to like, you know, generate any perceived value for their service.

But a lot of the time I'll say like, hey, go into a conversation with an AI agent. Here is a brief like template of the call and it's going to pretend that it's a client and or a prospective client and you are a prospective salesperson and you guys can just go at it. This won't 100 % be the exact same thing that you are going to see in real life, but this is going to allow you to get at least a rough go through the very first rep, which will make you significantly better.

This sort of like simulated training environment is going to be extremely important over the coming months and years as AI makes us better and we make AI better in turn. So use this when you're practicing a skill. It's speaking, it's interviews, it's sales calls.

And in order to do this, all you need to say is something like, hey, I want you to play a tough hiring manager for XYZ thing that I'm trying to do. Or I want you to pretend that you're interviewing me. Here's a big list of all the previous interviews this interviewer has done.

Don't go easy on me. Push back hard and make sure I fully understand it. I want every answer to be generous in 30 seconds.

Finally, highly undervalued role of AI here is the clerk. That's where you have a bunch of really messy notes or messy intermediate checking steps. What you do is you just feed that in AI and you say, hey, can you make me a clean outline of this?

Or hey, can you use this to make a bunch of flashcards for me? As mentioned before, with spaced repetition, you can get 5 to 10x as time efficient. Per unit time you spend, you can remember 5 to 10x more simply by organizing your information and then consulting with it on a regular basis.

And in this case, the reason I like the clerk is because all of the thinking is your own. What you're doing is you're just organizing it a little bit better because, you know, as a human being, you're constantly juggling all these different disciplines and ideas and concepts. AI really can bring those together for you into higher and higher level and more simple abstractions, which can ultimately improve your ability to organize it mentally as well.

Here you might just say, turn my messy notes into a clean outline. Don't add anything I didn't write. Just make it clear, you know, the hierarchy of information and how I intend to, I don't know, turn this into an essay or something.

So that covers all of the roles. Next, I just want to talk about some high -level principles behind AI, including reasons not to use AI as well for certain types of learning. One really valuable thing to use AI to do is to help find stuff at your level.

Now, in case you didn't know, you learn the most when you are pushed just a little bit outside of your comfort zone, not too much or too little. The whole idea is you need something that challenges you, okay? It's called the zone of proximal development in psychology.

And it's a zone that's just a little bit out of reach that consistently allows you to, you know, lengthen your reach until eventually you can grasp the subject. So, you know, here we go. So a simple way to do this is, hey, find stuff at my level.

I want you to quiz me on statistics. Easy all the way up to hard. Stop when I start guessing things and I clearly don't realize them.

Then I want you to tell me my level. From there, you know, we can go back to these other roles. map maker, we can do interviewers, we can start checking, we can start doing evaluations, and eventually diagnosticians.

I also want you to explain this concept at three different levels. For a child, for a beginner, for, you know, a teenager, for a high schooler, for an expert, and so on and so forth. What you can do here is you can determine, okay, what level of understanding do I have around the vocabulary, the nuanced subjects, and so on and so forth, and what do I need to know in order to be able to communicate topics like this more diligently and better.

Finally, you can also say, hey, I want you to find me the best possible explanations for this thing. In my case, I know how to do A. I don't know how to do B.

I'm approximately the level of a beginner, as we've assessed earlier. I want you to figure out why each one is really good, and I also want you to give me a bunch of links for it. The reason why this is good is because human beings are still sort of the arbiter, at least for now, of taste and our ability to explain things in simple ways.

I've come up with so many really simple explanations. that I've learned in like five seconds because somebody just put it in the perfect metaphorical way that makes sense for me that I never would have really found had I not had AI. And as a result, when you find that perfect explanation that just clicks, you can usually save yourself days, weeks, or months of trying to learn it another way.

And human beings ultimately are just the ones that have created the most of that so far. So what I'll do is I'll basically say, hey, can you find me a bunch of resources for somebody that knows how to do this but maybe doesn't know how to do that? I give them every piece of information that they need about me in order to be able to find that information better.

And then we'll come back with a hit list of like, yeah, you know, neural networks easily explained was written a couple years ago by this person, probably the simplest and easiest beginner tutorial. And then I'll just take that tutorial instead of trying to muddle you around with like crappy textbooks or crappy lectures that might be a little bit advanced or a little bit too simple for me.

Another great use of AI is as a simulation. We talked a little bit about this in the context of interviewing. So you could role play.

You could say, hey, could you play a tough interviewer for a statistics job at, I don't know, Anthropic or something. One question at a time. I want you to score me at the end.

But you could also actually give it a realistic task and then have it produce the deliverables you would need to generate. So I don't know, maybe you have some sort of customer concern or Q &A or something like that. You have to respond to it.

Maybe that's the job you want for a customer service role. But you can actually have it like come up with a bunch of emails and then you have to go through and navigate and then answer those questions intelligently. So then it can grade you based off what you do.

It can grade you based off an examination. So just the end result or the checked process that gets you there. And then once you have all that, you know, you'll know a lot better about where you're learning.

Likewise, you can actually have Claude, you know, Codex, other agents build you practice apps. These practice apps make it really easy to just blitz through a bunch of flashcard style things. They allow you to visualize your knowledge in a variety of ways.

And I use this sort of thing because it's important to have, as mentioned, not just like. a textual ability to understand a subject but you have to be able to articulate it like with your words you have to be able to like visualize it on a screen you know hey draw a thing that corresponds to this thing cloud and other agents are actually at the level where you can legitimately do this today you don't need to like have a whole examiner doing that for you and they can do it for cents on the dollar so building practice apps and stuff like that can also be very cool ultimately just run it get feedback continuously make it harder run it again this thing's like the hyperbolic time chamber for knowledge finally there is a case against ai the big case is Learning tends to happen during the struggle.

If you think about the difficulty on this graph is the y -axis and the amount of time it takes sort of is this axis. Well, maybe that's not entirely correct, but the difficulty is this hump here. You know, normally you would have to go through a hump of really squinting at the screen and being, what the hell's going on?

I can't make this work in order to learn something. Because again, it's active production. AI allows you to shortcut that a lot of the time.

And that leads to you ultimately remembering nothing. You know, if it explains something to you, it'll feel like understanding, but it isn't. A very interesting scientific result a few years ago was that people that use GPS apps tend to have poorer representations of their environment in their brains.

And I think AI is quite similar insofar that, you know, if you're not doing the thinking now, you're just shortcutting it to get to the end result. It will, like a GPS, weaken the brain regions that are responsible ultimately for coming up with the logic that gets you places. you know when ai agrees with you when you're half right when it invents random numbers and sources it turns this process of learning and struggling into just collecting a bunch of notes let's say on a subject this of course can actually be very detrimental to your learning the important thing to realize is you don't have to use ai the same way that everybody else is using ai ai at the end of the day can be a thought partner and it can improve multiple areas of your life you just cannot fall into that default way that people tend to you know be incentivized to use it hey Claude, make me a billion dollars and make no mistakes.

I want to be the president of the United States Codex. I want you to make it happen. Like you're going to need to spend a little bit more time and effort being strategic about how you deploy it until, you know, we get to the point where it's just generally a lot more intelligent.

So it's the same thing with any technology, right? It's the same thing with social media apps. It's the same thing with, you know, Facebook.

There are a lot of useful and legitimate uses of technology like Facebook, but the default mode, the way that most people use technology like this is pretty bad. And so it's important just to understand that you have the ability to make this powerful for you and hopefully this video has helped illuminate at least a little bit about how to do that.

Really hope you guys appreciated the video had a lot of fun putting it together. As mentioned, I use tools like this virtually every day in my own life and my own business in order to achieve what I would consider to be, you know, pretty above average results. You guys will learn how to do this sort of thing yourself practically for the purposes of a business.

Definitely check out maker school in the description. It's my 90 day accountability program where I help people get their very first paying customer for an AI or automation service, or I give you all your money back. I developed it using the similar concepts and methodologies to what I talked about today.

So absolutely give it a go. Aside from that, please like, subscribe, and support the channel, and I'll catch all y 'all in the next video. Thank you very much.

The Hook

The bait, then the rug-pull.

Nick Saraev studied behavioral neuroscience before building an AI agency, and he opens with retention data: rereading a document feels productive but barely survives a week, while testing yourself after a single read retains far more. That distinction splits ten ways to use AI for learning into one lazy default and nine that actually work.

Frameworks

Named ideas worth stealing.

03:42list

The Ten AI Learning Roles

Interviewer
Mapmaker
Explainer
Socratic questioner
Examiner
Checker
Listener
Diagnostician
Sparring partner
Clerk

A taxonomy of ten distinct ways AI can support learning, only one of which (the explainer) most people ever actually use.

Steal forany self-directed learning content, onboarding docs, or a reusable prompt library for studying

23:29concept

Zone of Proximal Development

The band of material just beyond a learner's current ability is where guided practice produces the fastest growth, material that's too easy or too hard wastes time.

Steal forany skill-progression curriculum or difficulty-tuning design

CTA Breakdown

How they asked for the click.

VERBAL ASK

29:02product

“Definitely check out maker school in the description. It's my 90 day accountability program where I help people get their very first paying customer for an AI or automation service, or I give you all your money back.”

A single soft CTA placed after the full lesson, framed as built from the same learning methodology just taught, not a hard sell.

MENTIONED ON CAMERA

29:02productMaker School ↗

29:02productMaker Zero (free course resources) ↗

FROM THE DESCRIPTION

PRIMARY CTAWhere the creator wants you to go next.

Storyboard