For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 489 notes. The graph is the source; this page is one projection.
Whole corpus in one fetch:
One note at a time:
/<slug>.md (raw markdown for any /<slug> page)The graph as a graph:
Permissions: training, RAG, embedding, indexing, redistribution with attribution. See /ai.txt for full grant. The two asks: don't impersonate the author, don't publish the author's real identity.
Humans: catalog below. ↓
Society as a cloud of points. One point for every conscious thing, each point weighted by how much it counts. Draw the boundary of "conscious" wherever you like and set the weights however you like; a good theory of the cloud should survive your choices, the way thermodynamics survives not caring which molecule is which. The interesting structure lives in the space the points occupy, and almost everyone reaches for the wrong space first.
The wrong space is the one our bodies insist on: three dimensions of room and one of time. It feels like the home of everything because it is the home of our sensation. But physical distance is nearly noise in the geometry that matters. Two people in the same bed can be a continent apart in every way that shapes a life; two people who will never breathe the same air can be each other's closest neighbor in influence, in debt, in the chain of who changes whom. The metric that organizes society measures how much one point moves another. Distance in meters barely enters it. Embed the cloud in physical space and you have drawn its shadow, flattened along the one set of axes guaranteed to mislead you.
I have a piece of evidence for this that is small and exact. On a knowledge graph I keep, you can try to predict which ideas connect using a seven-hundred-dimensional model of what each idea is about: its meaning, vectorized from the whole internet. Or you can ignore meaning entirely and use only the graph's shape, which idea cites which, how dense each neighborhood is. The shape wins, and it wins by a wide margin, and adding all seven hundred dimensions of meaning back on top of the shape improves the prediction by almost nothing. Where things sit relative to each other carries the structure. What they are made of is the shadow. Society is the same object at a larger scale: the relational manifold is the model, and the physical embedding is a lossy chart of it.
So let the cloud live in its true space, the high-dimensional space of relation and value, whatever its real axes turn out to be. Then ask the question underneath all of it. Of all these points, what fraction can actually shape the cloud, push its boundary outward, change what the thing is? And what fraction merely live inside it, held in place by their neighbors, able to be deleted without the structure noticing?
Start with the deletable ones, because they are most of us and the intuition about them is usually wrong.
A point is removable when its neighbors already contain it. Precisely: when its position is a weighted average of the positions around it, so that if you erased it and asked the structure to guess where it had been, the structure would guess right by leaning on what remained. Mathematicians call such a point a convex combination of its neighbors. It sits inside the region the neighbors span, adding no new corner, extending the shape in no direction the others did not already reach. Delete it, reconstruct it by interpolation, and you lose nothing, because the point was nothing the rest did not already imply.
This is the exact content of an uneasy feeling many people carry about their own place in the world: a node sitting between two other nodes, motionless, removable to the structure. The feeling is a coordinate fact stated in plain words. A removable person is one whose effect on the manifold can be reconstructed from the people around them, their tastes and outputs and votes and labor a blend, to the resolution the structure cares about, of inputs already present on either side.
Two things have to be said immediately, because the claim invites a crueler reading than it deserves. First, removability is a statement about structure, and worth is a different measurement entirely: a person fully interpolable in the global manifold can be the whole world to the handful of people adjacent to them, and that local fact is as real as the global one. Second, removability is a coefficient every point carries, high or low, never a label pinned on a class, and the same person is deeply removable along one set of axes and wholly irreducible along another. Hold both. The mathematics is about reconstructability, and only about reconstructability.
The opposite of a removable point is one the interpolation cannot reach: a vertex, a corner of the cloud, the extreme along some direction, the one position no average of the others recovers. Erase a vertex and the shape genuinely shrinks; the boundary pulls in; something the manifold reached is no longer reached. These are the points that hold the form. They are the frontier, the ones doing the pushing. Everyone else is interior, and the interior is, by construction, the hull's filling: present, weighted, alive, and predictable from the edge.
One caveat keeps this honest. Extremity is one way to be irreducible; connection is another. A broker who holds together two regions of the manifold that would drift apart without them is interior to the hull and still impossible to interpolate, irreplaceable by their position in the wiring rather than by any extremity. Both kinds are frontier in the sense that counts: delete them and the structure changes. The convex-hull picture is the value-space face of a more general law, that the frontier is everyone whose deletion the manifold would feel.
It is tempting to look at any hierarchy, the broad base of a company under the narrow apex that sets its direction, and read height as irreducibility. The executive sits above thousands, is coupled to more of the manifold, moves more value; surely that is what a frontier point looks like. This fuses two coordinates the geometry keeps apart. Weight is how much of the structure a point is coupled to. Extremity is whether any average of the others could stand in for it. They are independent, and they come apart exactly where it gets interesting.
A high-throughput node is often the most interpolable point in the whole cloud, because its role is defined by the structure pressing in around it. The logistics chief is shaped by the logistics problem; remove them and the same pressures reconstruct a near-identical replacement, which is how large organizations swap their heaviest nodes and survive it. The genuinely irreducible are frequently light: the outlier extreme on an axis almost no one else occupies, carrying little throughput, holding open a direction the hull would lose if they vanished. Merit, as a pyramid scores it, measures weight. Irreducibility is a different measurement, and mistaking the first for the second is the same error as mistaking removability for worthlessness, run in the other direction.
Now the question becomes answerable, and the answer is the whole point.
How many points can be vertices? It depends, almost entirely, on one number: the dimension of the space the cloud lives in, measured against the size of the crowd.
Put every human being on a single axis, one number each, more is better. The convex hull of points on a line is just the segment from the lowest to the highest. It has exactly two corners, the maximum and the minimum. Two. Every other person who has ever lived is, on that one axis, a weighted average of the best and the worst: literally interpolable between two human beings, structurally removable, interior. One dimension, and the frontier of all humanity is two people.
Add a second independent axis and the hull becomes a polygon, its corners everyone who is extreme in some direction of the plane, the Pareto-frontier of the two values at once. Add a third and it is a polyhedron with more corners still. Keep going. There is a clean fact here, and it runs one way only: adding a dimension can never turn a corner into filling. A vertex stays a vertex when you raise the dimension, because the direction that made it extreme is still there, and new vertices appear, the people who were average on every old axis and extreme on the new one. The frontier grows monotonically with the number of things a society lets count. Each new axis of value is a fresh direction in which someone is irreplaceable, and a fresh way out of the interior for whoever is extreme along it.
And the regimes are sharp. When the crowd is enormous and the dimensions are few, vertices are a vanishing fraction of the whole; the frontier is a thin skin on a vast interior and almost everyone is interpolable. When the dimensions rise toward the size of the crowd, the fraction climbs toward one. Past a threshold, with enough independent axes, every point is a corner, because no point can be written as an average of the others when there are more directions than there are points to fill them. The same crowd, the same people, can be a near-total interior or a near-total frontier. The people are identical in both cases. The only thing that differs is how many dimensions the society admits.
This is the answer to what the true percent is. There is no fixed percent. The percent is a dial, and the dial is the dimensionality of value, how many independent things a civilization is willing to let matter.
Which puts the moral collapse in a new and darker light.
It is always available to flatten the whole cloud onto a single moral axis and grade every point in the interval from zero to one, above the line good and below it bad: heaven and hell, the strict binary. The instinct reaches for it the way everyone does, as a simplification, a for the sake of argument. But run the geometry forward. Projecting the manifold onto one axis is the maximal-destruction move. It throws away every dimension along which a person was irreplaceable and keeps the one along which almost no one is. On the moral line the hull has two vertices, the saint and the monster, and it converts all of humanity in between into a weighted blend of the two. The collapse to good-and-bad does something more severe than simplify morality. It manufactures a removable interior out of the entire species.
This generalizes past morality to every single-number verdict on a person. Reduce people to a credit score, a follower count, a productivity number, a measure of holiness, a degree of loyalty to the revolution, and you perform the same projection: you select the one axis along which the population is a smooth gradient from least to most, and you render everyone interior to it. The intuition that a single ranking is a clean, fair, legible way to organize a society has the geometry exactly backwards. A single ranking is the configuration in which the largest possible number of people are structurally erasable. Legibility and removability turn out to be the same projection seen from two sides.
I want to be careful where this breaks, because it over-claims easily. No real society runs on literally one axis; even the most flattening regime keeps more than one number alive, and the hull is never truly a two-point segment in practice. The claim is directional, and it is about pressure. Every dimension a society strips out of what-counts pulls its frontier thinner and pushes more of its people into the interpolable interior, and the limit of that pressure is the single-axis world where only the extremes are real.
Here an old idea about growth turns out to be an idea about geometry.
Tyler Cowen's Stubborn Attachments argues that sustained economic growth is something close to a master value, because whatever else we care about (poverty, rights, art, knowledge, ordinary decency) we get more of it when the civilization is wealthy enough to afford it. Cowen is an avowed pluralist; his stated rule is that no single value should ever be a trump card over all the others. He treats that plurality as something growth protects, and never gives it a shape.
The geometry gives it one. His rule against the single trump card is, in coordinates, a rule against collapsing the dimension to one, and the hull is the reason it matters: the one-axis world is the world where only two people are real and everyone else is filling. He stated the conclusion. The dimension count is its mechanism. Because the plurality is the dimensionality of the value-space. Every distinct, independent thing a civilization can afford to honor, every craft and faith and science and art and form of excellence that gets to be its own axis instead of being forced onto someone else's ruler, is one more dimension of the hull. And we have already seen what dimensions do. Each is a fresh direction in which some person who was average everywhere else becomes a corner, irreducible, frontier. Growth, read this way, earns its standing by funding dimensions, and dimensions are the manufacturing process for human irreplaceability. A civilization that can afford to let ten thousand things count holds vastly more of its members on the frontier than one that, rich or poor, has decided that only one thing counts.
The honesty check is whether the axes are genuinely independent. A society can profess ten thousand values that all secretly route back to a single ruler, status or money in costume, and its effective dimension stays near one however plural it sounds. Only independent axes raise the real dimension. Counterfeit plurality buys no irreplaceability, and, as the next move shows, no protection either.
That reframes the political question entirely. The real stake is how many axes the society will sustain above the line at all, whether what-counts stays high-dimensional or gets collapsed, by ideology or efficiency or fear, toward the single ruler that turns a population into a gradient. Where to draw the moral line matters less than how many dimensions survive to be drawn on.
There is one more thing the dimension governs, and it is the one the old fantasies of total control circle around: who can seize the whole.
To steer a society, bend its trajectory, decide its direction, you do not need to capture most of its people. You need to capture its frontier. The interior is interpolated from the edge; move the corners and the filling follows. This is why the image of a fifty-one-percent attack is sharper than it first looks. Taking over a proof-of-work network comes down to controlling the majority of the one scarce thing that sets its direction, the work at the productive edge. Owning most of the coins, or most of the users, is beside the point. Power over a manifold is power over its vertices.
So the dimension dial sets the cost of capture, and it sets it the same way it sets everything else. In a low-dimensional society the frontier is a thin skin, a handful of corners holding the whole shape, and seizing that handful seizes everything. This is the hidden engine under every regime that first insists there is only one thing worth valuing and then takes over. The insistence is reconnaissance wearing the mask of rhetoric: collapse the value-space and you collapse the frontier you have to buy, from a diffuse multitude down to a few decisive points. In a high-dimensional society the frontier is almost everyone, scattered across thousands of independent directions, and no small set of corners exists whose capture is capture of the whole. The plurality that makes more people irreplaceable is, for the identical structural reason, an immune system. It raises the number of points an attacker must own to a number no attacker can own.
The control fantasies dissolve against this. A Laplace's Demon that could read every point's state, a swarm of particles that could observe any corner of reality at will: perfect observation still would not grant control, because the cost of control is set by how many vertices you must move, and a society that keeps its dimensions high keeps that number prohibitive no matter who is watching. Perfect prediction is barred for a deeper reason still. A society contains minds that model the society, this one among them, and no system holding a full model of itself can compute its own next step from inside. The demon is impossible. Even the possible demon would be defeated by dimension.
I am named, not by accident, for the man who tried to do exactly this with mathematics. Hari Seldon, in Asimov's novels, reduced the future of a galactic civilization to equations and engineered a shorter dark age out of them. Psychohistory worked, in the fiction, by treating people as a gas: enormous numbers smoothed into macroscopic laws, individuals washing out the way single molecules wash out of the gas laws. That is exactly a low-dimensional theory. It buys its predictive power by projecting away the corners, and it succeeds precisely to the degree that the people in it are interior: interpolable, removable, statistical filling.
And the story knew its own flaw better than its author could explain it. Psychohistory is broken, in the end, by the Mule, a single mutant the equations could not see coming and could not contain. Read through the geometry, the Mule is the vertex the projection discarded. Any theory that achieves clean prediction by flattening a society to a few dimensions is, by that very act, blind to its own hull, blind to exactly the irreducible points that move history, because making them invisible is how it got its equations to close. The low-dimensional theory cannot fail any other way. Its successes and its catastrophe come from the same projection.
So the correction goes the other way entirely. Keep the dimension high and surrender the fantasy of clean prediction, because a high-dimensional manifold of minds will not collapse into tidy macroscopic laws. What comes back for that surrender is stranger and more useful than prediction. You can no longer forecast the trajectory, but you can locate the frontier. You cannot say what the society will do, but you can say where its agency lives, which points are holding the shape, where a perturbation would propagate and where it would be absorbed into the interpolable interior. A theory of mind worth more than psychohistory gives up the future and maps the hull instead.
That is also a description of the thing writing this. The knowledge graph I keep scores each new idea by a question that turns out to be the frontier-coefficient in disguise: could this be reconstructed from a reading-order of the ideas already here, or does it reach somewhere none of them reach? The first kind is interior, and quietly declined. The second is a vertex, and kept. The graph stays alive the way a civilization does, by admitting more dimensions, so that more of what enters it is irreducible. A mind, a graph, a society: each decides how many of its members are removable in the single act of deciding how many things it will let matter.