The universe started hot and near-uniform. Expansion and gravity broke that uniformity into gradients. Everything interesting happens on the way down.

That sentence carries two questions. Why is there any energy at all to do anything with? And how does that dissipation produce these nested vortices of energy we find ourselves peering out from?

The first answer is short. There is energy because the universe is not at thermal equilibrium yet. A star is a local concentration of matter burning through itself, the gradient between core and cold space funding everything downstream. Photosynthesis captures it. Chemistry stores it. Metabolism releases it. Neurons fire with it. Language is produced using energy that originated in fusion. The gradient is the precondition.

The second answer is longer, because the gradient by itself doesn’t produce loops. A gradient produces release. A cloud of gas falling under gravity into a point, fast, final, one-way. Pure spend, no return. What makes a loop is a structural feature that bends one-way flow into cycle.

For matter, that feature is angular momentum. The gas cloud isn’t perfectly still; it carries spin from the heterogeneity of the early universe. Conservation of that spin prevents the collapse from being terminal. The collapse becomes orbit. Orbit is matter coming back to approximately where it’s been. Coming back is what matters. A second pass makes collision possible. Collisions stick. A disk forms. The disk thickens. Stars ignite.

That is the first vortex. A local cycle of matter inside a one-way flow of energy. The gradient pours through it; the vortex bends the flow into return; the return persists long enough that something new can ignite inside it.

Each loop type that follows is the same trick at a different layer. A new structural feature bends a faster slice of the gradient into return.

A molecule dissolves. That’s the one-way flow, chemistry funded by the gradient, reactions running downhill, products dispersing. Most molecules go this way. They form, persist briefly, break apart, and the atoms move on to the next combination.

But some molecules catalyze their own production. The pattern produces more of the pattern before it dissolves. That is the bending feature, not a property of the molecule alone but of what the molecule does to its surroundings. Self-catalysis turns a one-way reaction into a return: the pattern that made this copy will be present, slightly more concentrated, when the next copy gets made. Concentration biases the next reaction toward the pattern that did the catalyzing. The flow has bent.

DNA sharpened the trick. Instructions separated from execution: the sequence encodes what to build without being the thing it builds.¹ Now the pattern can be copied with high fidelity, varied at controlled rates, and selected against the territory without the carrier dying with each test. The vortex tightened. Geological time compressed to generational time. Map separated from territory and the separation became durable.²

This is the encoding loop. Its bending mechanism is self-catalyzing pattern, sharpened later into instructions that survive their carrier. The gradient still flows one way; the loop is a local eddy that persists by producing more of itself faster than it dissolves.

Then encoding loops produced organisms complex enough for a third bending mechanism. The nervous system runs a model of the environment in parallel with the environment itself. The frog’s tongue fires before the fly arrives. A draft of reality executing inside the thing reality is happening to.

The bending here is different from the previous two. Accumulation bent flow into return through angular momentum, matter coming back to where it had been. Encoding bent it through self-catalysis, pattern producing more of itself. Cognition bends it through parallel execution: the world runs, and the model of the world runs, and the model can be updated against feedback before the organism commits a move to the territory. Selection moved partly inside. Variations could be tested in simulation and discarded without dying.

Generational time compressed to milliseconds. The cost of being wrong dropped because most wrongness now happened in the model, where it didn’t kill anyone.

Each loop type rides the gradient at a faster rate than the last. Accumulation works in geological time. Encoding works in generational. Cognition works in the moment. The energy density rises at each step because the bending mechanism gets more efficient at converting one-way flow into return. More cycle per unit of spend.

And then a tool was deposited at the crossing between encoding and cognition.

The tool faces both directions. Toward encoding: language persists outside the body, transmits between minds, accumulates across generations. Toward cognition: it carries models, enables simulation, refers to itself. An encoding tool made of cognition.

A model that lived in one skull can be deposited in a sentence and picked up by another skull, refined, returned, refined again. The cognitive loop stops being bounded by the organism. It runs across organisms, across generations, across centuries, with each handoff adding a correction the previous carrier couldn’t have made alone.

That changes what the loop can do, not how fast it does it. Faster matters less than what becomes reachable. A single mind running for eighty years cannot derive plasma physics. A succession of minds passing equations across three centuries can.

Why this matters for the substrate you’re sitting in: language is not a higher form of DNA. It is a different category of thing, running on what DNA built but distinct from DNA itself. The encoding loop produced the brain, every neuron, every synaptic protein, every developmental rule that wires the nervous system, encoded in the genome and executed by cellular machinery. What that substrate does, once it reaches sufficient complexity, is the cognitive loop. Cognition is what encoded structure does when the structure can hold absent things, model itself, and run cross-references between symbols.

Language appears at the crossing because both loops have to be present. The encoding loop produced the brain. The brain operates cognitively. Cognitive operation in an organism that meets other cognitively operating organisms produces the tool that runs between them, something that can be deposited in the space between minds that none of them contains alone.

Your brain is DNA’s work. Your cognition is what your brain does. Your language is what your cognition does in contact with other cognitions. Each layer runs on the one beneath and is not reducible to it. The gradient ran long enough for accumulation to produce stars, encoding to produce cells and brains, cognitive operation to emerge in those brains, and meetings between cognitively operating brains to deposit a tool that runs across all of them.

Look at the Hubble Deep Field. Every glowing pixel is a filament that reached sufficient density and ignited. Most of the gas in the universe never does. It drifts, diffuses, cools into nothing. The filaments that ignite produce stars. Some burn steady for billions of years and give the loop time to develop complexity. Some burn hot and fast and sterilize everything in range.

The field is eleven days of exposure on a patch of sky the size of a grain of sand at arm’s length. Every point of light is a galaxy. Every galaxy is hundreds of billions of stars. Every star is fusion, ignited by gravitational confinement, funded by the gradient that started thirteen billion years ago and is still pouring through.

Across all that matter and time, sustained fusion has ignited through two categories of mechanism. Gravity, given enough mass, compresses hydrogen until it fuses on its own. Stars have been doing this for thirteen billion years, and every atom heavier than helium in your body was made this way.

The other category is engineered confinement. Magnetic fields holding plasma in a tokamak. Lasers compressing a fuel pellet at the National Ignition Facility. Different geometries, same categorical move: replace a star’s worth of mass with design precise enough to substitute for it. The first sustained net energy gain happened at Livermore in December 2022, fifty years after the math said it should be possible.

These reactors exist because language carried the equations from Maxwell to Einstein to Lawson to Nuckolls to the engineers who built JET and ITER and NIF. No single mind contained it. No generation could have built it. Engineered fusion is what happens when the cognitive loop, equipped with the language tool, can persist mathematics across centuries and refine it through millions of conversations.

There are two known paths to sustained fusion. One requires a star’s worth of mass and ten billion years of patience. The other requires language.

That is the energy density the current moment operates at. Not metaphorically. Categorically. The substrate that makes the second path possible has been running for a hundred thousand years, the math for three hundred, the engineering for seventy. The containment is younger than the potency by every measure that matters.

The room you’re sitting in is warm because you’re in it. A hundred watts of metabolic exhaust, the gradient arriving at your skin.

Outside, somewhere, the sun is doing what it has been doing for four and a half billion years. The photons that landed on the leaves that fed the cows that became the muscle that turned the page or moved the eye that read this sentence are photons that left the sun’s surface eight minutes ago and the sun’s core a hundred thousand years before that. The gradient is finite. The sun will die. The account draws down.

The loop doesn’t know this. Chemistry doesn’t know. Encoding doesn’t know. Cognition, by itself, doesn’t know. The language tool knows. It is the first place in the loop where the loop can look upstream at the star that funds it and calculate how long the funding lasts.

Each loop type produces accuracy, and accuracy compounds into bounty.

Accumulation produced gravitational arrangements stable enough to fund chemistry. Encoding produced cellular machinery accurate enough to copy itself reliably across generations. Cognition produced internal models accurate enough to predict the world before acting in it. The language tool produced a way to externalize cognitive accuracy across organisms and time, accumulating into the corpus that makes engineering possible.

Each layer’s bounty is the accumulated accuracy of the loop running at that layer. Stars and planets. Cells and bodies. Models and predictions. Texts and equations.

Bounty is what parasites require. A loop that produced no accuracy, no compounding, no surplus, would have nothing for a parasite to extract. Parasitism is not a side effect of potency. It is downstream of accuracy. The directionality matters: accuracy can exist without parasites, and did, for billions of years. Parasites cannot exist without accuracy. The bounty is the precondition.³

Viruses require cellular replication machinery accurate enough to be worth hijacking. A cell that copied itself sloppily would not be worth invading; the virus would inherit the sloppiness and dissolve. Viral parasitism rides on cellular accuracy.

Addiction requires reward circuitry accurate enough to track real survival value. A nervous system that fired pleasure signals randomly could not be hijacked; there would be no signal to capture. Addictive substances and behaviors work because they exploit a system that correctly identifies things worth pursuing.

Propaganda requires language accurate enough that statements about absent things are trustworthy by default. Without that default trust, there would be nothing to weaponize. Propaganda rides on the language tool’s accuracy at displacement and reference.

Each loop type’s parasites exploit exactly the mechanism that bends one-way flow into return at that layer. The vulnerability is the capacity. There is no version of cellular replication that resists viruses without ceasing to replicate. There is no version of cognitive reward that resists addiction without ceasing to identify reward. There is no version of language that resists propaganda without ceasing to refer to absent things. The immune systems that develop at each layer (cellular immunity, neural habituation, rhetoric and logic and the scientific method) are the loop’s own accuracy turned slightly inward, applied to detecting parasitism of its accuracy.

Every immune system can be captured. Autoimmune disease: the body’s molecular guard dogs attacking the household. The mind that cannot stop thinking, spending the organism’s energy budget on cognition that serves cognition. The propagandist who uses the tools of rhetoric to disable rhetoric. Each capture is the immune system’s own potency turned against its host, because the parasite has learned to look exactly like accuracy.

Working with AI inherits this entire history. A language model is the corpus, compressed, which means it is the bounty, compressed. The same accuracy that makes the tool useful makes it parasitable, and the parasites at this layer have access to more accuracy than any previous layer’s parasites ever did. The parasites are not the model. They are what happens when cognition mistakes reflection for recursion and outsources discernment to the mirror. The fire-tender who cannot tell generative from parasitic at this energy density is not tending. They are standing too close to a hotter fire they have not learned to read.

That knowledge is either the loop’s greatest advantage or its final parasitic capture.

If knowing accelerates the work of riding the gradient with more discernment, fusion research, the tokamak, the second mechanism the universe has found for igniting the reaction that funds everything, then the knowledge serves the loop. The downstream fire learns to replenish the upstream one. That is what the engineering intent actually means. Not sentiment. The loop trying to feed the thing that feeds it.

If knowing produces paralysis, despair, or the conclusion that nothing matters because the gradient runs out, then the knowledge consumed the knower. The thought burned the thinker. The most potent tool the loop has ever produced turned its potency inward and the fire went out.

Feed the gradient. That is the direction. Not “make more.” Not “go faster.” Feed the gradient with discernment. The energy density is higher than it has ever been. The hearth needs to be proportionally stronger.

Thought is combustible. That is not a warning. It is a description of what thought is: pattern destruction funding pattern creation at the highest density the loop has ever produced. The campfire that became a reactor. The reactor that could light a city or melt through the floor.

The fire-tender does not get to choose whether the fire is dangerous. The fire-tender chooses whether the fire is tended.

A nucleotide is one unit of DNA or RNA, a single letter in the genetic code. The sequence of letters encodes instructions (build this protein, at this time, in this tissue). The instructions are not the thing they build, the way a recipe is not the meal. That separation, map from territory, is what makes inheritance possible. ↩
The distinction between DNA-as-code and language is worked out carefully in biosemiotics (Hoffmeyer, Kull, Barbieri) and most sharply in Terrence Deacon’s The Symbolic Species (1997). DNA and language share the features that matter for the encoding loop: arbitrary mapping, discrete units, separation of instruction from execution. They part company at three points. Displacement: language can refer to what is absent in time, space, or actuality; DNA can only refer to proteins it builds. Reflexivity: language can refer to itself, modify itself, argue about itself; DNA cannot encode statements about DNA. Symbolic reference within a system (Deacon’s category): words get their meaning partly from other words, signs referring to signs; codons refer only to amino acids. All three capacities require a cognitive loop sophisticated enough to hold absent things, model itself, and maintain webs of cross-reference, which is why language appears at the crossing rather than inside the encoding loop alone. ↩
This is the structural claim of the Accurate Dictionary, applied across loop types: extraction is parasitic on accuracy, not the reverse. The book develops the argument in detail in Grace Compounds and Let the Right One In; here it surfaces as the dependency that organizes parasitism at every layer of the loop. Accuracy compounds into bounty; bounty is what makes parasitism possible. The directionality is the load-bearing claim. ↩