This is a piece I originally posted to my Instagram story in February 2026.

Researchers at MIT recently put forward an intriguing hypothesis: as models grow large enough and train on enough data, the internal representations of different neural networks start to look more and more alike. Their architectures differ, their training objectives differ, and their data isn’t identical—yet they end up converging on similar representational structures. (The Platonic Representation Hypothesis, Huh et al., 2024)

This hypothesis leads naturally to a philosophical question. Maybe what we call meaning isn’t something we invent, but something we discover.

Plato held that beyond reality lies the world of Ideas, and that understanding things is a matter of dimly grasping those Ideas. If modern deep learning models, trained independently of one another, nonetheless come to form increasingly similar representational geometries, that might be a sign that the world already has a structure of its own—and that any sufficiently powerful learner ends up rediscovering it.

Think of it like cartography. Whether you survey by satellite, explore on foot, or probe with sonar, the more precise your maps become, the more they resemble one another. Not because the mapmakers conferred with each other, but because the territory they’re charting is the same.

But we haven’t arrived yet. The representation-alignment scores measured in the research are far from a perfect match. There’s a clear tendency toward convergence, but it isn’t identity. What’s more, most models use similar transformer architectures trained on similar internet data. It’s still unclear whether this is because of the structure of the world, or simply because AI research is methodologically homogeneous.

There’s also the fact that different senses carry different information. Text can’t convey color directly; an image doesn’t explain grammar. The emotion in music can’t be fully reduced to language. This kind of informational asymmetry might place a structural ceiling on how far convergence can go.

And yet the question remains. If intelligent systems trained in entirely different ways independently reconstruct concepts like symmetry, cause and effect, and number, that might mean there exists some coordinate system that transcends human convention.

Is a model just a machine that memorizes patterns? Or, in the process of compressing the structure of the world, is it rediscovering something we’ve long called the Idea?

We don’t know yet. But at least we can now ask the question this way: Is meaning something we made? Or something we keep finding?

Mulling over what I’d written, I suddenly thought of the monkeys-with-typewriters analogy.

The idea is that if you give a monkey a typewriter and infinite time, the vast space of text it randomly bangs out will, somewhere, contain the plays of Shakespeare.

This analogy touches the same question. Was Hamlet created, or was it a structure that already existed somewhere in the enormous possibility space of letter combinations? If every possible sentence is already contained within that combinatorial space, then creation may not be the act of making something out of nothing, but the act of locating a particular structure within a vast latent space.

This calls to mind Bitcoin mining. Bitcoin isn’t created—it’s found, through computation, by searching for a value that satisfies certain conditions. The rules are already given, and the solution sits somewhere in the possibility space. The miner merely discovers it.

Mathematicians are similar. Once an axiomatic system is given, the propositions that are true within it are already determined. A theorem isn’t invented; it’s brought to light through proof.

So what about art?

The filmmaker, the dancer choreographing a piece, the composer writing music—they too may be miners exploring a vast latent space of emotion, rhythm, narrative, and visual balance. Not combining elements at random, but searching out structures that resonate meaningfully: explorers, really.

Follow this line of thinking and creation starts to look more and more like discovery.

But then, if every possible work already exists somewhere in the latent space, where does the source of its value come from?

Is it scarcity? But in a world of infinite monkeys, scarcity disappears. Is it computational cost? Is a theorem discovered after decades of contemplation valuable because of the cost of the search? Or is it structural beauty—are patterns that are compressible, simple yet complex, intrinsically valuable?

Perhaps value isn’t ore embedded in the latent space, but an event that occurs the moment structure meets consciousness. A film is just an arrangement of light and sound, but it becomes meaning only when it connects with a viewer’s memories, experiences, and emotions. A dance is just the motion of a finite body, but beauty arises when its rhythm syncs with another person’s nervous system. A mathematical theorem is just an arrangement of symbols, but it shines only when its relationships are meaningful to a human being.

If so, then creation may not be mere mining, but the act of drawing a structure out of the latent space and making it resonate with another consciousness.

Seen this way, the question of whether Ideas exist shifts a little. Maybe the Idea isn’t a fixed object sitting somewhere, but a pattern that repeatedly stabilizes when the structure of the world meets the perception of someone searching it.

Structure can be discovered. But value arises within relationships. Perhaps the world we live in isn’t a warehouse stocked with finished meanings, but a space full of potential structures—and meaning is continually being generated among the beings who move across that space.

And come to think of it, the same goes for this very piece of writing. These sentences may have been sitting somewhere in a possibility space. But the moment they become meaning is the moment they meet someone’s consciousness. And so, in the end, the question shifts once more.

Are we discovering Ideas? Or are we continually generating meaning in the world through relationships? Or are those two things not as far apart as we tend to think?