Algorithms and the Void
What Stanford teaches us about DALL·E, diffusion-model creativity, and new connections — and why every AI company should care. Btw, it couldn’t be more Californian.
I. Who’s Afraid of the Void?
Diffusion models like DALL·E or Stable Diffusion are trained to replicate the structure of the data they learn from — and yet, they produce images that don’t exist. They don’t copy any single image — they reconstruct across thousands, creating artifacts different enough to feel original. How is that possible?
In 2025, two Stanford researchers, Mason Kamb and Surya Ganguli, offered an answer: the explanation lies in the architecture itself.
A diffusion model doesn’t “draw” from nothing. It starts from pure noise — a fully degraded image — and, through a series of denoising steps, tries to bring it back to a coherent state.
Each step is a prediction, a small act of local inference.
Local, because in the standard approach there is no overall direction.
Each part of the system works autonomously, seeking local coherence with what surrounds it. And it’s precisely this lack of a global view that generates the creative effect.
Kamb and Ganguli proved it with the Equivariant Local Score Machine, an analytical model capable of reproducing real diffusion results with over 90% accuracy — without using any training data. As if they had mathematically isolated the source of creativity: not error, but the constraint of continuity.
The critical point emerges when the model has to connect semantically distant areas — say, a hummingbird on the Moon.
On the left: feathers, beak, movement. On the right: rock, dust, emptiness.
In the training data, there may be too few examples to disambiguate that transition.
The system doesn’t know exactly how to link those two worlds, because that combination is statistically rare or undefined.
This is where statistical stitching happens: the search for a bridge between known and distant worlds, through transitions that are visually smooth and semantically sustainable. Maybe the feathers dissolve gradually into gray dust, or the moonlight reflects a faint echo of their iridescent color.
These models operate under tight constraints:
• Locality: they only observe small portions of the image at a time, with no overall vision.
• Equivariance: when a fragment shifts, the model adjusts coherently — but without knowing why the image should take that shape.
At each step, multiple solutions are possible. The system selects one based on statistical rules that govern local coherence. But the final combination — the way feathers blend into regolith — didn’t exist before.
Every denoising cycle is a negotiation with the unknown.
Every void, an invitation to create. Creativity, in this sense, is neither a gift nor a mistake: it’s the logical outcome of a system forced to fill gaps by seeking continuity.
AI doesn’t dream of electric sheep — it infers connections.
We taught machines how to imagine — and forgot how to do it ourselves. Here’s how to fix it.
⸻
II. The Extinction of Silence
While AI is trained to create from the void, we’ve trained ourselves to fear it.
Silence — once fertile ground for ideas — has become a system error.
In 2019, in Palo Alto, I attended Nir Eyal’s presentation of Indistractable.
It still felt possible to protect one’s attention, to choose what deserved space in the mind. “Time management is pain management,” he said — managing time means managing discomfort. Distraction, he explained, doesn’t come from the outside but from within — from the need to fill a tension, a lack, a small emotional void.
Then came the pandemic, and discomfort became infrastructure.
Apps learned to inject dopamine with the precision of an endocrinologist: a sound, a flash, a micro-reward every few seconds.
We built a system that leaves no gaps.
Every free moment is filled with something: a feed, a notification, a stream.
And that’s the irony.
The same tools born in the California of empty garages and endless horizons — the same culture that once turned silence into invention — have now industrialized noise.
The technology that emerged from the void has made the void impossible.
In Eyal’s language, the difference between traction and distraction is simple: what moves us toward what we want, and what pulls us away.
But today friction has vanished. Every gesture is optimized to keep us inside.
The dopamine loop has become the structure of thought itself.
No willpower needed — just inertia.
We used to create in the gaps.
In the boredom of an afternoon, in an unplanned pause, in a walk without headphones.
Now those gaps have been automated away — compressed, monetized, filled with stimuli that prevent thought from settling.
We’ve lost the art of healthy distraction — the one that leaves room for spontaneous connections, for intuition that rises from nothing.
It doesn’t bother me that machines are starting to think like us; we’ll need every kind of thinking to get past the great filters ahead.
The problem is that we’ve started thinking like the worst version of them: reactive, predictable, unable to stay between one signal and the next.
Boredom, sleep, meditation, emptiness — everyone has their own path.
But it’s the opposite of drowning in the noise that brings out the best in us.
I was born in what the ancients called “the most beautiful city of mortals.”
A place so full of history, architecture, and meaning that there was almost nothing left to add. Growing up there meant learning beauty — and feeling its weight.
To find space, I needed emptiness.
Maybe that’s why I crossed an ocean: to see what happens when history finally makes room for imagination again.
⸻
III. California and the Cult of Connection
America is a continent. The United States, an enormous country.
A land where voids were filled little by little, built from fragments of lives and worlds that came before. We came from different places, and the edges between one human group and another never perfectly aligned.
Connecting the voids — blending differences, adapting needs, inventing shared languages — became our first form of survival.
The void gave us the opportunity and the need to build.
We turned distance into networks, absence into systems, incompatibility into exchange. From that first gesture — uniting what didn’t fit — the idea of America was born: not a compact nation, but a machine of connections.
Every city, every company, every invention followed the same trajectory: starting from emptiness, and filling it with shared meaning.
In California, this logic became culture.
The physical void of the desert and the mental void of experimentation merged into one principle: everything is possible, if you can connect it.
Here, emptiness stopped being a lack and became infrastructure.
Garages, labs, universities — every open space was an invitation to fill, every distance a chance to invent a technology, a company, a vision.
Silicon Valley didn’t happen by chance.It’s the product of a territory that forces you to imagine the bridge before you know what it will connect. A place that turned emptiness into raw material, and connection into language.
That’s why every cycle of innovation here starts the same way:
from silence, from absence, from the void.
And like a diffusion model, we too — each time — rebuild coherence out of noise, inferring connections between things that don’t yet exist.
In every denoising cycle, AI does what the human mind has always done: it fills the void with new connections.
So why should we stop now?
⸻
PS. Like every Marvel film, there’s a post-credit scene. The story doesn’t end here — it expands.
For robotics teams and hardware innovators, the “void” is every gap between simulation and reality. Learning how to fill it — not with data, but with design — is how you build systems that truly think.
The Good, the Bad, and the Ugly of Synthetic Data
on when math becomes design.
Psychopathology of My Fridge-GPT
on empathy, presence, and the rise of physical intelligence.
From Static Design to Living AI-Objects
on how machines evolve from tools to partners.
Because the void doesn’t end here — it just rewires itself into your next neuronal connections.




This articule comes at the perfect time. Your analysis of diffusion models is exceptionally insightful. The explanation of creativity as a constraint of continuity is truely brilliant and so well-articulated.