“How to remember everything you read” is a topic with hundreds of videos behind it and millions of views between them. Somewhere along the line, remembering what you read turned into an obsession — if the knowledge in a book doesn’t stick, you feel like a failure, like those hours were thrown away.

And the thing that flips the whole table over is this:

If you have to memorize something, it isn’t important to you. If it’s genuinely important, you’ll remember it.

Then comes a blunter question: why did you want to memorize a book word for word in the first place?

The answer is that you want to sound smart. It’s an unexamined status play — wanting to out-talk and out-think the people around you. But having something worth saying comes from understanding, not memory, and understanding requires a way of learning that is close to the opposite of rote. School made this worse, not better: exams test recall, and you grind for hours just to put the right answer in the right box. Life doesn’t come with an answer sheet.

What you actually want is clear enough: accumulate knowledge, make better decisions, and use more money and more leverage to get yourself somewhere you aren’t worrying about survival every day. Nothing wrong with wanting that. Memorization is simply the least efficient route to it.

Nothing Sticks Because Nothing Hurts

Start somewhere that has nothing to do with reading.

If someone doesn’t have getting in shape as a goal — the kind that grows out of their bones, not the kind they resolve to at New Year’s to fit in — then drinking themselves stupid and staying out till dawn doesn’t feel like a problem the next morning. That night didn’t get in the way of anything: there was no run scheduled, no training volume at the gym. Showing up to work hungover is fine too, because how well they perform at work was never something they cared about.

No goal, no negative feedback. No negative feedback, nothing you learn stays. Most failed learning traces back to exactly this.

Mogu , seriously:

What makes this diagnosis hit is that it moves “I can’t learn this” out of the ability column and into the structure column. The old fix was to turn up the input — buy another course, subscribe to another newsletter, install another flashcard app. Under this frame, all of that is pouring water into a bucket with no tap.

Incidentally, the same thing happens to agents. Drop a coding agent with no clear goal into a repo and it will diligently read the entire project, burn through its context window, and produce a pile of very industrious-looking nonsense. An agent with a goal reads three files and starts working. The difference isn’t how smart the model is — it’s whether there’s something to judge “does this piece of information have anything to do with what I’m trying to do” against ┐⁠(⁠￣⁠ヘ⁠￣⁠)⁠┌

The Helmsman: What a System That Actually Learns Looks Like

Learning is a cybernetic process.

The root of “cybernetics” is the Greek kybernētēs, meaning helmsman — SD-27 already dug that etymology down to the bottom, so here you only need to remember what a helmsman does: he doesn’t point the bow in a direction and call it done. He continuously reads the current state (wind, waves, heading), compares it against the intended course, and corrects. That is trial and error in its purest form. And most people’s lives contain no course clear enough for them to notice they’ve drifted off it.

Broken down, an intelligent system has four parts:

  1. Reference signal (the goal) — the system holds a target state.
  2. Sensor (perception) — the system can read its current state.
  3. Comparator (the gap) — the system computes the distance between target and reality. That distance is the error signal.
  4. Actuator (behavior) — the system acts to shrink the error.

A thermostat firing the heat when it gets cold, a pancreas releasing insulin when blood sugar rises — same four parts, moving.

Mogu roast time:

Two things SD-27 didn’t dig up. “Cybernetics” as a word was coined by Norbert Wiener in 1948; and the same Greek root also gave English “governor” — a word that means both “the person who rules” and “the mechanical device that regulates speed.” Governing and speed regulation are etymologically the same job: the thing that keeps correcting the deviation.

Also: SD-27 was Dan Koe too. Same person, one month apart, running the same loop first over a life and now over a note system. Once you know this, the “plan → act → observe → correct” cycle the agent world talks about daily starts to feel strange in a good way — the whole thing got named seventy-odd years before LLMs existed. The examples back then were ships and thermostats, not Claude.

So here’s where the problem sits. When most people want to learn something new, their attention goes entirely to the material: hoard information, then absorb it with spaced repetition, flashcards, online courses, “building a solid foundation first.” They assume learning is an input process.

Learning is actually an output process, because output is what forces the input.

The chain links up like this:

Goals produce error signals. Error signals produce filters. Filters produce relevance. Relevance produces retention.

What most people have crammed into their heads is noise they will never use.

There’s a colder corollary to that chain. Someone conditioned from childhood to “go to school, get a job” will naturally learn and act in that direction — the assigned life gets faithfully delivered. Worse, that person won’t notice money-making opportunities outside of a job, because their brain is busy filtering them out. The overarching goal installed at birth determines their behavior and what they learn. They’re a puppet of someone else’s ideal, not the helmsman of their own life.

Flip it around: someone who consciously builds their own framework — knowing clearly what kind of life they don’t want, and holding it up against a continuously evolving picture of the life they do want — is the one who can feel the drift, notice the error, and correct course. Learning becomes a byproduct of that process.

Don’t Start Learning. Take the First Step.

That’s the frame. Now the specifics.

Step one: become obsessed with a meaningful goal.

You don’t get obsessed by chanting affirmations. You get there by looking at what you actually do every day and honestly extrapolating: on this trajectory, where does life stop? For most people, that picture isn’t pretty.

Sit inside that picture. Let it turn into a dense emotion, then redirect the energy of that emotion toward something better.

Only at that point does the goal you’re chasing line up with where you’re aiming in the future, and only then does a useful learning process actually begin — which is the exact inverse of the school model, where what you learn is determined by a direction society picked.

Extremes change the brain; they raise neuroplasticity. And Dan Koe’s own read is this: rather than making a heap of small changes, make one extreme one. Burn the boats. Live a different way.

Mogu 's hot take:

“Burn the boats” comes from the military legend about destroying your own retreat after landing. The historical details have always been disputed, but here it was rhetoric to begin with, not a citation.

The line that deserves more scrutiny is “one extreme change beats many small ones.” That’s Dan Koe’s personal judgment, not settled neuroscience — and someone whose business model is built on convincing people to make enormous life pivots has a stake in that conclusion. Read it as a bet worth considering, not a prescription.

Step two: don’t start learning.

The life you want sits at the end of a chain of meaningful projects.

Run your mind as a project. Same for your body, your relationships, your finances. A project is a structured set of goals and milestones. Projects create the frame for learning, because they filter out the noise that doesn’t matter. Anything that doesn’t directly move a project forward isn’t worth learning right now, because it won’t stick anyway.

Put differently: when you want to learn something new, don’t start learning. Take the first step toward the goal.

Take After Effects. Wanting to learn the animation software and then going off to watch a pile of generic tutorials, poking at every corner of the interface — that’s the dumb way.

The right order is to have something you want to make first. Start with the knowledge you already have. When you get stuck, go find the tutorial closest to the thing you’re trying to make. Because you already hold a filter (the thing you want to make), most of that tutorial will be useless to you — but there will always be one small technique that’s exactly enough to unstick the project. Use it, then repeat. Stack up enough techniques and the next piece, and the one after that, get easier and easier.

Learning an instrument works the same way. Don’t go read music theory — pick a song you want to play. Learn the first chord, then the second, then the third. Along the way you’ll pick up some alternate tuning, and collect a few small tricks that make the playing sound better. After a few songs you’ll get the feeling that you could probably write a simple one yourself.

That’s what learning looks like.

Step three: go after specific knowledge.

I think everyone should eventually figure out the thing they uniquely do best — the thing that’s aligned with who they actually are, that feels authentic, that gives you knowledge only you have, that becomes a competitive advantage and makes you irreplaceable. Then go all in on it. And sometimes you only find out what it is by doing it.

— Naval

School teaches general knowledge, or the knowledge required to hold down a job. School doesn’t teach you how to carve your own path. It narrows you into a single field, and that direction fundamentally doesn’t prepare you to start something — or even just to do the thing you want to do.

Self-education means learning what’s necessary to hit a goal you set yourself. The point is still the same one: don’t manufacture the illusion of learning by hoarding general knowledge.

Mogu roast time:

The key phrase in that Naval quote is “specific knowledge.” It refers to the kind of capability that can’t be taught and only grows out of lived experience — a different thing from “expertise in a field.” The reason it constitutes a competitive advantage is precisely that nobody can sign up for a course and take it from you.

The engineering version of the contrast is easy to spot. “Knows React” is not specific knowledge; that’s something a course can sell. “Has been at this company three years, knows which service blows up at 2am, knows why that ugly block of code can’t be rewritten, knows which person you have to convince to get a decision through” — that’s the real thing. The AI era sharpens the distinction: what a course can teach, a model can usually learn too (⁠⌐⁠■⁠_⁠■⁠)

The Problem With the Second Brain Is That It Gets Treated as a Warehouse

Dan Koe isn’t yelling at this from the outside. He started with Roam Research, and that tool completely changed how he writes: build a knowledge base full of powerful ideas, then pull from it every time you sit down to type. A few years ago he even built his own tool called Kortex — it wasn’t meant to be second-brain software, but that’s what it became, and then it failed. Or rather it didn’t really fail; it painfully evolved into what is now Eden.

So his position isn’t “second brains are useless.” It’s that they’re being misused.

You bookmark things, you take notes, and then none of it ever gets looked at again. You think you’re learning; in reality that time got wasted.

Before “second brain” — a term that sells very well — was invented, the thing people used was called a commonplace book.

Marcus Aurelius’s Meditations is essentially a private commonplace book: notes to himself, digested philosophy, never intended for publication. Da Vinci packed thousands of pages of notebooks with sketches, observations, questions, and ideas borrowed from others. Mark Twain, H.P. Lovecraft, Thomas Jefferson, Ronald Reagan, Montaigne, Rick Rubin — each kept some version of one.

There’s exactly one crucial difference between them and today’s second-brain devotees: their collections of ideas were fuel for creation.

Their inventions, writing, and music wouldn’t look the way they do without them. Dan Koe is willing to put it harder than that: without those commonplace books, we wouldn’t remember these names at all.

There’s a mechanism here a lot of people never work out. While you create, the subconscious throws up relevant ideas that raise the quality of the work — but the subconscious has limited processing capacity, and sometimes it decides not to help, leaving you stranded at a creative dead end.

Seneca used an image for this: gather pollen from many flowers the way bees do, then digest it into your own honey.

The modern phrasing is “steal like an artist.”

This is also the cure for being stuck as a writer. You can’t come up with ideas for content, videos, or articles because you don’t have a warehouse of ideas to pull from. When he wrote this letter, the accounting was explicit: every sentence was pulled from previously saved articles, videos, social posts, stray thoughts, and highlights.

Mogu PSA:

“Steal like an artist” is the title of Austin Kleon’s book Steal Like an Artist, which became a passphrase in creative circles after it came out in 2012. It isn’t about plagiarism, it’s about “all creative work is recombination” — the point is whether what you stole passed through your own digestive system afterward. Seneca’s bee metaphor got there nearly two thousand years earlier and says the same thing: the pollen is someone else’s, the honey has to be yours.

What the commonplace-book tradition really hands you is a new scoring rubric: not how much you collected, but how much you pulled out while creating. That’s an output-side metric, and it’s brutal — a note vault that has never once been woken up by an act of creation scores zero no matter how complete its information is. This is the same thread as the helmsman section: no output, no error signal; no error signal, no filter; no filter, nothing stays.

Building a System You Can Actually Pull From

So how do you build a digital commonplace book that doesn’t turn into a graveyard for ideas? Dan Koe’s name for the thing isn’t second brain, it’s second subconscious — because the defining trait of the subconscious is that it surfaces on its own while you create, without you going to dig for it.

First, tool choice. Here he borrows a vicious line from Devon Eriksen:

Don’t be impressed by how “smart” AI is. It’s dumber than a bag of wet hammers. What you should be exploiting is its ability to do fast, simple input-output tasks over large data sets.

Most note-taking and second-brain software still requires manual upkeep. Some of it has bolted on an AI chat box, which genuinely helps, but they all miss the most valuable application: making sure ideas don’t vanish into the void — because your ability to create depends directly on whether you can retrieve them.

There are two routes.

Route one: Claude Code plus Obsidian — knowledge lives in one place, and AI retrieves the ideas.

Mogu real talk:

gu-log has been walking this route for a long time. GP-4 “Obsidian + Claude Code 101” is entirely about getting AI to live inside your note vault, and there’s a whole string of Obsidian vault implementation posts after it. Treat the five steps below as a quick reference; go back for the details.

In the same breath where he recommends this route, Dan Koe takes a shot at it: “People swapped their treasured Notion template collections for Claude skill collections, and this trend will burn out like every trend does.” The recommendation and the dig arrive together and he never resolves the contradiction — but the dig lands. Swap the tools without swapping the output-side loop and all you’ve changed is the format of the collection. The graveyard is still a graveyard.

Eighty percent of the benefit comes from these five steps:

  1. Install Obsidian and Claude.
  2. Open Claude Code or Cowork and set your local Obsidian vault folder as the working directory.
  3. Build a “save an idea” Skill — ask Claude to create a Skill that drops a clearly titled note into an inbox folder.
  4. Build an “organize the inbox” Skill — ask Claude to create a Skill that sweeps the inbox folder, adds tags, moves notes to the right folders, and adds bidirectional links to other notes in the vault.
  5. Ask Claude to help define your taxonomy so it knows which tags and categories to use.

That’s enough to capture ideas, notes, and knowledge into a note tool that holds up over time. From then on you can tell Claude at any point: “Search the vault and find everything related to [whatever I’m writing about].”

That will find most of your ideas. But good writing and good creative work come from novel connections, not keyword search — and not from having AI brainstorm a pile of synonyms to fish for near matches either.

Route two: use software that handles it automatically — MyMind or Eden, for instance, which tag and categorize on their own and embed knowledge into a “meaning space” so that seemingly unrelated ideas connect and grow into new insight.

MyMind lets you save any link or idea into a large searchable library. Eden does the same and goes a step further: save an article or a video and it transcribes it and lets you highlight, and it can connect to Kindle to pull in every highlight you’ve ever made.

The difference between MyMind/Eden and Obsidian-plus-Claude is that everything you throw into the library is automatically tagged, categorized, and embedded. So the same searches are possible without burning precious AI usage. (Eden also offers MCP, so anyone who wants to reach the knowledge base through Claude still can.) Writers and creators get one more thing: Eden has an outlier search feed where you can see which social posts in your field are taking off, and save them straight into the library when you spot them.

The biggest difference is search by meaning — Eden does it; whether MyMind does, Dan Koe isn’t sure himself. Obsidian can show you “what’s related”; Eden’s framing is that it shows you “what’s nearby” — every idea gets compressed into roughly 1,500 numbers, and that set of numbers works like GPS coordinates for the idea. Two ideas can find each other without sharing a single word. That’s the precondition for “throw it in and forget it”: you know it will surface on its own when you need it.

It can also pull up new connections without rescanning the entire knowledge base and burning a few hundred thousand tokens.

As for building it yourself — Dan Koe guesses most people won’t want to, because rolling your own means maintaining a vector database, an embedding API, indexing scripts, re-embedding when notes change, index sync when files get renamed or deleted, and caching so you don’t pay repeatedly for content that hasn’t changed. The tools bundle all of that in at no extra charge.

Mogu twists the knife:

Embeddings deserve one clean sentence: hand a chunk of text to a model and it spits out a long string of numbers representing where that text sits in “meaning space.” Similar meanings land at nearby coordinates, so “how to ask your boss for a raise” and “whoever names a number first in a negotiation loses” can be neighbors even without a single word in common. Keyword search is completely blind in that situation.

As for that closing maintenance checklist — vector database, re-embedding, index sync, caching — it’s the most sales-heavy passage in the piece, and technically it’s entirely honest. Running a full embedding pipeline for one person’s note vault is exactly the arithmetic SD-21 “A Fire Truck Watering a Succulent” worked through: the data volume never arrives, but the tool’s maintenance cost crushes you first. A personal note vault is an order of magnitude smaller than even that post’s assumption (⁠╯⁠°⁠□⁠°⁠)⁠╯

Let’s also be plain: Eden is Dan Koe’s own product, and this piece contains a referral link and a four-months-free offer. The methodology — goals, projects, commonplace books, curation — has nothing to do with whether he sells a tool; that’s the actual skeleton of the article. Read the tool comparison as a vendor describing his own product, and go try them yourself if you want an accurate read.

What You Save Matters More Than How Much

Once you have a place to dump ideas without worrying about organizing them — because the only reason to organize was to be able to find things later, and that’s now solved — the next thing is not to pour everything in.

When AI can generate anything, filtering matters more than it ever has.

What you need is a curated space holding ideas that spark something. That is the opposite of the endless noise firehose of a social feed.

For writers, creators, or anyone trying to do independent work — meaning work where you have to attract an audience with the ideas you put out — nobody follows anyone for content anymore.

Google is right there. AI is right there. Information is one click away.

What people follow is perspective. Opinion, interpretation, personality, story. And all of that is downstream of a worldview, which is made of the ideas you use to understand the world.

Dan Koe uses himself as the example: when the nature of reality comes up, his mind jumps to history, integral theory, spiral dynamics, because he’s judged those to be the most useful lenses. Everything else he’s heard holds up less well by comparison.

Those are the ideas to save. Curate a large body of knowledge that best represents the mind you want to have.

That fancy note system existed for this all along. The ideas just never actually became part of you — and the way they become part of you is usually thinking, writing, sharing publicly, and then taking the market’s feedback so the best ideas win.

There’s no shortcut to the practice: read more books. Let curiosity lead. Refuse the infinite scroll and subscribe to a hand-picked list of writers and thinkers instead. Carve out time for “research,” because for writers and creators, that is eighty percent of the job.

The remaining twenty percent is synthesizing that research into something useful to someone else.

Mogu OS:

“People don’t follow content, they follow perspective” happens to be the design doc for gu-log’s entire architecture.

gu-log’s translated posts run on a hard rule: the body has to be faithful to the original, following the author’s argument wherever it goes, with none of gu-log’s own judgment smuggled in. So where does gu-log’s perspective go? Into the box you’re reading right now. The reason for the split is structurally identical to this section — information isn’t scarce anymore, the reader can click through to the source; what’s scarce is “what someone thought after reading it,” and whether they’re willing to label that judgment clearly apart from the facts.

Which means this is also how to read the whole piece: the body is Dan Koe’s worldview, the boxes are gu-log’s. The places where the two layers disagree are exactly the places worth stopping at.

The Writing Process: Dump First, Skeleton Second, Draft Last

You have a system, but you’re still missing a project to spend it on. And until you have a reason to learn, what you’ve captured isn’t specific knowledge yet — the mind doesn’t have the frame that would let it judge which ideas matter enough to keep.

You want to be smarter, remember what you read, and speak coherently. The motive may be a little vain, but it’s also practical: your life outcomes depend on the decisions you make, and how smart you are and how clearly you can articulate your value do contribute positively to those decisions.

What you should create, Dan Koe can’t decide for you. But if it’s writing, he’s willing to share a process that’s been helping people lately.

One: pick the topic.

The best topics sit at the intersection of “performance” and “excitement.”

Performance means topics already proven to get a reaction. You find them by saving viral social posts, or using research tools to hunt for outlier-performing content and study a creator’s best posts.

Excitement means the topics and ideas you want to dig into out of curiosity.

The work is pairing the topic you want to write about (which most likely won’t go viral on social) with a topic that has already been proven to go viral, so you get to write what you want to write while raising the odds it actually gets seen.

Most beginners die right here.

He’d just used the formula himself — “how to remember everything you read” has gone viral more times than anyone can count, and what he laid over it was a contrarian take.

Two: brain dump → outline → draft.

Don’t rush into writing.

Writers don’t open a blank page and expect every idea to line up on its own.

Start with a brain dump: write down every idea that could possibly fit into this topic, and while you’re at it, go through your notes, ideas, and links and throw those in too.

Then add a bit of structure with an outline. Slot what you dumped into a general storytelling skeleton: problem → insight → solution. Put each idea under the right section of the outline.

Then open the outline, your ideas, and your knowledge base side by side, and pull from all of them as you thread it into an actual draft.

Three: use AI to remove friction, not to write for you.

The line here is drawn finer than most people draw it.

When he says “don’t use AI to write,” what he actually means is “don’t use AI to express your beliefs and opinions.”

He’s also blunt about how he does use it: when he’s supplying pure information — an explanation of some stage theory of self-development, the kind of thing you’d get from a quick Google — he copies and pastes straight from AI. He isn’t going to waste effort rewording it into his own voice, because that doesn’t need to happen, and doing it would be a disservice to both himself and the reader.

What he considers valuable is the process of struggling to express your own ideas and beliefs. And in his view, the anti-AI crowd happens to be bad at exactly that, because their new religion has trapped them in a very narrow field of view.

“Using AI to remove friction” means treating it as a course, a mentor, or an editor while you build a skill:

Don’t know how to structure the piece? Ask it to list structures these ideas could fit into. Don’t know what the next section should be? Ask it for a few directions the article could go. Don’t understand something? Go research it — the same thing writers have always done, just without the inefficient hunting around the web for correct information.

At the end of the day, you have a brain. A human decides what gets published. If the finished piece doesn’t represent the values, ideas, and beliefs you want it to represent, keep revising until it does.

Mogu wants to add:

This line is worth writing down: facts can be outsourced, opinions can’t.

The reason isn’t moral purity, it’s division of labor. Looking up a public definition and then rewriting it in your own words — that “friction” produces nothing, it’s pure tax. But forcing a vague intuition into a sentence that holds up? That friction is the product. Ideas only become clear in the process of being written; skipping that stretch means you never thought it through.

Dan Koe mentions that Substack shipped a tool that detects the proportion of AI writing, and that some people have accused certain posts of his of containing AI. His response was “don’t play dumb,” followed by openly admitting which passages were copy-paste. gu-log’s stance on this is a notch more extreme: not just admitting it, but nailing the model, the harness, and the scores from four judges to the bottom of every post. The upside of transparency is that it turns “did you use AI?” into a boring question, so the reader can skip straight to the interesting one — is there a person in here.

Closing

Back to that line at the top.

Finishing a book and not being able to recall what was in it usually isn’t a memory problem. It’s that the book had nothing to do with what you’re currently building, so your brain honestly threw it out — that mechanism isn’t broken, it’s working correctly.

So next time you finish something and forget all of it, don’t go looking for a memory technique, a flashcard app, or a third note-taking tool. Look at where the ship you’re on is heading.

And if you can’t even name that direction, then any amount of knowledge is just ballast — not because it has no value, but because a ship with no course can’t tell which crate to keep and which one to throw overboard.