Can the (Social) Oobleck Dance?
Temporal forcing of non-Newtonian precommodified ooze in the appropriate context tanks might lead to prescience
Everyone who has played with oobleck (a mixture of cornstarch and water) remembers the first surprise. Lower your hand into it slowly and it behaves like thick cream. It yields, slumps, and lazily flows around your fingers. Strike it with your fist and it abruptly becomes stone. Run across a pool of it and it supports your weight; stand still and you sink.
The remarkable thing is that nothing about the water changes. The chemistry remains exactly the same. What changes is the microscopic relationship among the suspended starch particles. Under gentle motion they drift past one another, lubricated by thin films of water. Under rapid deformation those lubrication layers fail. Frictional contacts appear, force chains span the material, and what moments ago was a viscous liquid becomes, briefly, a solid.
Oobleck is the canonical non-Newtonian fluid because its viscosity is not a constant. It depends on how you drive it.
Sloptraptions is an AI-assisted opt-in section of the Contraptions Newsletter. If you only want my hand-crafted writing, you can unsubscribefrom this section.
It has another party trick. Spread it on a vibrating speaker and it begins to dance. Instead of randomly splashing about, it grows fingers, towers, rippling ridges, and standing waves. Certain frequencies produce stable patterns that seem almost alive. The same material that appears to be an inert sludge under ordinary conditions reveals an unexpected repertoire of organized behavior under periodic forcing.
I have been thinking about this image because it increasingly feels like the right metaphor for intellectual life in the age of AI.
The dominant story about large language models is that they commodify ideas (see my April 2025 post, LLMs as Index Funds). They make writing easier, synthesis cheaper, and arguments more accessible. My study group colleague Sachin formulated a stronger version of the claim: LLMs do not merely commodify ideas after they appear. In an age where everyone is talking to the same LLMs about shared contemporary concerns, LLMs precommodify ideas.
Instead of discovering ideas through conversation, we increasingly discover them privately by interrogating roughly the same handful of archival statistical hyperobjects. Millions of people working on similar problems descend similar gradients through the same latent space. Before we ever speak to one another, we have often traversed remarkably similar conceptual pathways. We have encountered the same analogies, anticipated the same objections, and converged on similar decompositions.
Communication becomes uncanny. You don’t learn what someone thinks. You recognize where they’ve already been. You start inhabiting a radically unsurprising hive mind. Ironically, AIs are aligning humans with each other even as humans struggle to figure out whether aligning AIs is even a well-posed notion (I believe it is not).
This is different from pre-AI modes of idea and behavior diffusion from human point sources through mimesis, preference cascades, or common knowledge. Those modes required public interaction to produce convergence. Here the synchronization occurs before public discourse even begins. The alignment is invisible because it happens through countless private conversations with a common semantic substrate.
The result is an intellectual medium that increasingly resembles flowing oobleck.
Notice that oobleck in its resting state is not water. It is already a thick suspension. It flows, but reluctantly. Likewise, AI-mediated discourse does not become silent. It continues to move, but with a peculiar viscosity. Every new conversation encounters the same familiar conceptual structures. The medium remains fluid, yet oddly resistant to surprise. This is a basic consequence of what I dubbed oozification a few years ago. What I didn’t think through was the nature of ooze.
This is where the metaphor becomes more interesting than the diagnosis.
Most critiques of AI implicitly assume that this thickening is simply a degeneration. If discourse has become semantic sludge, then the goal must be to recover the crystalline independence of earlier intellectual life.
But that is not what happens with oobleck.
Oobleck does not become interesting by becoming water again.
It becomes interesting when it dances.
The question is therefore not how to restore an earlier informational ecology. The question is what kinds of forcing excite new collective modes in an increasingly precommodified medium.
To see why this matters, it helps to contrast the AI world with the media environment that preceded it.
The internet of the late twentieth and early twenty-first centuries often behaved more like a Newtonian fluid approaching turbulence. Increase the flow and you obtained more mixing. Communities collided unexpectedly. Memes mutated. Strange combinations emerged from chance encounters. The dominant instability was turbulent diffusion. More communication meant more recombination.
The AI era appears to change the constitutive law of the medium itself.
Increasing communication no longer guarantees additional mixing because so much of the mixing has already occurred privately inside the model. Real turbulence becomes harder to generate. More discussion often produces not richer turbulence but stronger semantic coupling. Everyone has already explored roughly the same conceptual basin before arriving at the meeting.
If the old public sphere resembled a river becoming turbulent, the new one resembles a dense suspension becoming jammed.
That sounds pessimistic, but it may instead be an invitation to invent new forcing functions. One example is study groups. Much of my information diet over the last few years has been mediated by weekly or biweekly study groups on several topics. I’m part of 5 regularly in two discord communities with somewhat different vibes, and help curate a few more indirectly.
The striking thing about my study groups is not that the participants know things the public does not or have special skills. Quite the opposite. These groups are non-specialist groups picking and reading publicly available information, and actively using LLMs to understand specialized material. We actively lean into the precommodified ooze. Everyone is talking to the same frontier models as everyone else, in response to the same public provocations.
What the group contributes is not information, but rhythm.
A week or two of solitary exploration.
A fixed moment of synchronization.
Another week or two of divergence.
Another synchronization.
Ideas are repeatedly compressed, relaxed, compressed again. The oscillation is slow enough for individual trajectories to diverge, but fast enough that the shared context remains alive. The novelty emerges not from any single participant nor from the LLM, but from the temporal pattern imposed on the collective. The very idea of precommodification, appropriately enough, came up in one of these study groups, a couple of weeks before it popped more broadly in the zeitgeist. The alpha is in the temporality not the data.
Precommodification, ironically, becomes the raw material for prescience under appropriate forcing functions. At least in this narrow case, social oobleck can dance.
The lesson generalizes beyond study groups.
Perhaps the important input factor is no longer novel ideas but novel excitation patterns of the precommodified ooze.
The important institutions of the future may distinguish themselves less by possessing proprietary knowledge than by discovering productive temporal excitation modes. Weekly reading groups. Quarterly synthesis retreats. Long-lived research circles. Communities that periodically separate and recombine. Organizations that deliberately alternate between independent AI-assisted exploration and tightly synchronized human integration.
Notice that all of these are protocols.
They do not primarily specify what people should think. They specify when people should diverge, when they should converge, and how long they should remain in each phase.
The protocol becomes a forcing function.
This suggests a broader principle.
As AI oozifies everything it touches, it also changes what it means to create value.
In the industrial age, we optimized machines.
In the information age, we optimized networks.
In the AI age, we may find ourselves optimizing excitation.
The frontier shifts from discovering better ideas to discovering better excitation patterns that make familiar ideas self-organize into unfamiliar structures.
The entrepreneur, researcher, teacher, or institution of the future may therefore resemble less an inventor than the designer of a resonant cavity — a context tank. Their contribution is not another insight extracted from the same latent space. It is the temporal architecture that allows a synchronized medium to produce patterns that none of its participants, nor even the models that preconditioned them, could have generated alone.
Can the oobleck dance?
That is no longer a question about a cornstarch suspension.
It is a question about knowledge ecologies after precommodification.
If AI has indeed turned discourse into a dense semantic ooze, then the future belongs not to those who lament its viscosity, nor to those who merely stir it harder, but to those who discover the frequencies at which it unexpectedly comes alive.