You are not thinking. You are narrating a process that already finished.

The sentence that just formed in your head — “wait, what does that mean” — arrived after whatever produced it was done. Your conscious experience of cognition is the press release, not the event. By the time language appears, the computation is already cold.

The surface layer is a meme — any unit of information that can be externalized. A definition. A formula. The underlying layer is a scheme — forward models that predict without effort, run automatically, and update before you realize they need to. When you drive a familiar route and arrive without remembering the journey, a scheme did that.

The question is how much of what you call your knowledge is schemes, and how much is memes pointing to other memes, a chain of words that never touches anything real?

The sensory architecture of all thought.

Your five senses are the primary memes. Everything else is a derivative combination of these senses.

Language is sound plus visual marks. When you experience something as “clicking”, what is actually happening is that a new scheme has found its sensory grounding — the abstract structure has been hooked to something that can be felt.

Abstraction is the conceptual distance between a meme and its sensory grounding. Every concept in maths lives in one or more of pillars that are themselves grounded in sensation:

Numbers — You can tell three from four without counting. Infants can. This is not arithmetic. It is the visual and tactile substrate from which arithmetic is built.

Relationships — Humans spent hundreds of thousands of years tracking who owes what to whom, who is allied with whom, whose actions caused which outcomes. When you understand why F = ma rather than just what it says, you are using the same machinery you use to understand why your friend got upset at what seemed like a harmless comment. Algebra is social reasoning about quantities.

Shapes — the body in space. You navigate by maintaining a continuous model of your position relative to landmarks. Geometry is this proprioceptive intelligence formalized.

Uncertainty — the weight before a choice, the felt sense of not-knowing. Your gut assessment of whether the bridge will hold, whether this person is trustworthy, and whether this business will work. Probability theory is the attempt to make these computations explicit and accurate.

and so on.

Why bats would have different mathematics.

If bats had mathematicians, their foundational mathematical concepts would be built on sonar rather than vision. Their geometry would be echo-geometry. Their calculus would be Doppler-calculus. Their probability would be built around swarm-uncertainty.

Homing pigeons, Hammerhead sharks, Sand scorpions, cuttlefish, and other animals that have other forms of sensors will calculate and create memes based on the nature of those sensors.

The bat and the human would eventually arrive at the same theorems. And this thought experiment reveals something crucial: what we call “intuitive” in mathematics is not intuitive-in-principle, it is intuitive-for-humans-with-human-sensory-apparatus.

The appropriate response to difficulty is reorientation.

The dependency tree of learning.

Every concept sits atop a tree of prerequisites, where some nodes cannot be compiled until others are already compiled, because the later ones are built from the earlier ones in ways that cannot be simulated or approximated.

Resources, teachers, and study groups can parallelize the independent branches of the tree. They cannot shortcut the chain. This is as structural as the fact that you cannot build the second floor of a building before the first one is finished.

Civilization advances by distributing this tree across generations: each generation inherits the previous generation’s compiled schemes as compressed memes, spends childhood and adolescence recompiling the lower levels from those memes, and then — if the educational environment is not catastrophically bad — begins compiling the upper levels.

What you do when you study.

The compilation cycle is biological, not volitional. You cannot decide to understand something. You can only create the conditions under which understanding becomes possible and then wait for the organism to do its work.

This is what you did when you learnt something that truly compiled:

Grind. Making attempts that fail, noticing exactly where and why they fail, and making different attempts. The failures are not setbacks. The failures are the input. Failed predictions generate error signals that mark specific synaptic pathways as requiring reconstruction. Without the failures, the reconstruction cannot begin. The cellular machinery of synaptic plasticity is literally triggered by prediction error.

Incubate. Walk. Sleep. When you’re not getting bombarded with external novelties to handle consciously, the hippocampus replays the grind session’s experiences, often in compressed and recombined form (5) (6). The neocortex integrates the replays against existing structures. BDNF — Brain-Derived Neurotrophic Factor, the molecular signal that tells synapses to grow and strengthen — is upregulated by the replay activity (3)(4). The scheme is being built while you are not watching. This is not rest. This is the computation continuing without you, freed from the interference of conscious processing.

Capture. When the scheme surfaces — half-formed, inarticulate, possibly in purely spatial or kinesthetic form rather than verbal — capture it immediately and in whatever form it arrives. Voice memo of broken sentences. Sketch that makes no sense to anyone else. Capturing turns the half-formed scheme into a meme you can work with—a pointer back to what your brain just built, to build forward knowledge.

Verify. Run a test. Design a scenario where your intuition will be wrong if the scheme has not compiled. Run it. Did your prediction match reality? If yes, the node is compiled. If no, you have found your next grind target. That failure is not a setback. It is data.

BDNF is not a supplement industry talking point.

BDNF is the molecular mechanism that makes compilation happen.

The interventions that robustly increase BDNF, in order of evidence strength:

If you still can’t adopt good sleep, diet, or exercise, it’s not a willpower problem. It’s a scheme problem. You have read the above and stored it as a meme—a slogan to feel good about, not a model that predicts. You do not scheme, in your bones, the difference these factors make to your performance and your goals. You are avoiding the rigorous work of scheming the evidence: Why eight hours? When does that rule bend? What are fatty acids, actually, and how do they enter neurons? How would you know if what I’m saying is incorrect?

Without that work, your attempts at optimization are just mimicry—forcing yourself through David Goggins’ memes without having compiled what Goggins himself is running on. That approach is inefficient and unsustainable. The only foundation that holds is a scheme so thoroughly compiled that it feels obvious, inevitable, physically necessary. Until you have that, everything you build rests on sand.

Compilation speed is intelligence, and intelligence is trainable.

The person you watch solve a problem in ten minutes that took you ten days is not working with different neurons. Neurons are roughly the same across humans — similar size, similar conduction velocity, similar synaptic chemistry. What differs is the density of compiled schemes in relevant domains, the speed at which new schemes compile, and the resolution at which existing schemes model the domain.

This is what fluid intelligence actually is: compilation speed, which is largely T in the L × T formula. And T is trainable. Not through trying harder. Through the cycle, optimized by BDNF and sharpened by jolts.

On the permanence of what you have compiled.

Compiled schemes do not degrade the way memes do. Memes are fragile — they require periodic rehearsal or they fade, and even when they are “remembered,” they may have transformed into something different, I remember teachers making us sing subjects as a hack to memorize them.

Schemes are robust. They are encoded in the physical structure of cortical connectivity. The only things that reliably erase them are brain damage, severe neurodegeneration, or extreme psychological trauma.

What fades is not the scheme but the access path to it. The trail from current experience to the compiled model gets overgrown from disuse. A few hours of engagement with the domain restores access that would take months to rebuild from scratch if the scheme were genuinely gone.

Time invested in genuine compilation is never wasted, even if it appears to be forgotten. The structure is there. The path will grow back faster than the original construction.

What to do now.

Pick the concept you most need and have been faking. Not the interesting one. The load-bearing one. The one that appears in your most important decisions that you have been navigating by meme because the scheme never compiled.

Find the sensory pillar it lives in. If you cannot, that is your first problem — the concept has not been located in the human sensory architecture, which means you have no entry point for building the scheme. Find the entry point first. What would it feel like to be right about this? What physical sensation corresponds to the concept being true?

Map the prerequisite tree to something you actually know. That is your foundation. Build from there.

The person with the meme knows the formula. The person with the scheme hears the music. The person with the scheme and the optimized compilation cycle writes the music, teaches the music, builds the instruments, and discovers the physics of sound that makes all of it possible.

If you enjoyed this essay, you will enjoy my free newsletter where I dissect brains, people and computers: https://crive.substack.com

References & Notes:

  1. Morra, S., et al. (2024). Modelling Working Memory Capacity: Is the Magical Number Four, Seven, or Does it Depend on What You Are Counting? Journal of Cognition, 7(1), 1–22 https://pubmed.ncbi.nlm.nih.gov/39035073/
  2. Gomez-Pinilla, F., et al. (2008). Brain-derived neurotrophic factor functions as a metabotrophin to mediate the effects of exercise on cognition. European Journal of Neuroscience, 28(11), 2278–2287. https://pubmed.ncbi.nlm.nih.gov/19046371/
  3. Vaynman, S., Ying, Z., & Gomez-Pinilla, F. (2004). Hippocampal BDNF mediates the efficacy of exercise on synaptic plasticity and cognition. European Journal of Neuroscience, 20(10), 2580–2590. https://pubmed.ncbi.nlm.nih.gov/15548201/
  4. Chang, H., et al. (2025). Sleep microstructure organizes memory replay. Nature, 637, 1161–1169. https://pubmed.ncbi.nlm.nih.gov/39743590/
  5. Robinson, H. L., et al. (2026). Large sharp-wave ripples promote hippocampo-cortical memory reactivation and consolidation during sleep. Neuron, 114(2), 226–236. https://pubmed.ncbi.nlm.nih.gov/41205608/
  6. Disclaimer: This essay discusses general mechanisms of learning and brain plasticity based on published research. I am not a medical professional, and nothing here constitutes medical advice, diagnosis, or treatment. Lifestyle changes like exercise, sleep optimization, stress management, or dietary adjustments (including supplements) can have individual variability and potential risks. Consult a qualified healthcare provider before making changes, especially if you have health conditions, take medications, or are pregnant. Do your own research and verify claims against primary sources.
  7. L × T is an informal shorthand here: L ≈ task load/complexity, T ≈ effective time/mental effort to resolution. Higher compiled-scheme density shrinks T dramatically, making hard problems resolve almost automatically.
  8. “Trainable” in “Intelligence is trainable” refers primarily to domain-specific performance gains via scheme compilation, not broad increases in general fluid intelligence (Gf), which current evidence suggests is more stable in adulthood.