No. 001
Monte Carlo Daydreams
A simulation does not predict the world. It rolls a die a million times and watches what shapes the dust forms. Every outcome a brushstroke; every distribution a watercolor wash.
monte-carlo
A pop-art gallery of simulation AI
Pinned to the wall, painted in watercolor, wired with circuits. Walk the gallery of simulated minds.
Each card is a pinned experiment. Some are watercolored, some are halftoned, some are wired. Read them in any order.
A simulation does not predict the world. It rolls a die a million times and watches what shapes the dust forms. Every outcome a brushstroke; every distribution a watercolor wash.
Bayesian updates rendered in Ben-Day dots. Each new observation darkens a swatch of the prior, the posterior a printed page slowly resolving from noise to picture.
A simulated mind is not magic. It is wires and gates and feedback. Below: a tiny circuit drawing itself when you scroll it into view.
in → gate → feedback → out
“ All models are wrong, but some of them POP.
— gallery placard, gallery 2
A storm modeled in a laptop has no thunder, but it rains the same hard arithmetic. We paint the gradient and the lightning in the same warm orange.
Run the model a hundred times. Print the results on the wall. Notice the sameness, then the difference. The grid below repeats one prompt with one nudge each pass.
"Boss, we built a simulation of the storm."
"Did the model rain on it?"
"POP! It rained ten thousand times."
Watch the circuit ink itself onto the page. Inputs flow left. Feedback returns. The gold path is the loop that learns.
The gold path is what makes a simulation a learning simulation.
Eclectic pieces from the back room: agent-based fables, fluid-dynamics doodles, neural-net portraits.
One thousand tiny agents on a hand-drawn village square, each following three rules. Civilization, or its watercolor approximation, emerges by sundown.
A small feedforward portrait. Watch the neurons connect like a courthouse circuit diagram drawn in a notebook from 1973.
3 → 3 → 2 // dense
Smoke through a vent rendered in halftone. Coarse-grained on the inside, swirling on the surface. The math is finite-volume; the feeling is finger-paint.
"What if Tuesday had rained?" The simulator answers in three frames: before, nudge, after. The third frame always feels uncanny.
“ A simulation is a question you can ask ten thousand times while the kettle boils.
— placard, gallery 3
A wall of dice rolling at once. Halftone for the static, watercolor for the trends. The wall is loud and the average is quiet.
"Run it again." (she said, calmly.)
"Run it nine thousand more."
"POP! Now we know what Tuesday means."
Penciled into the margins of the gallery: parameters, observations, gotchas.
[run] seed = 1717 steps = 10_000 warmup = 500 [noise] shape = "gaussian" sigma = 0.18
"After 8,000 rolls the histogram is no longer noisy — it is a watercolor. Print it."
Don't trust a single run. Don't trust a single artist. Run a thousand of either; then trust the average.
[t= 0.00s] init ok [t= 0.04s] warm ok [t= 3.21s] step 1000 [t= 6.39s] step 2000 [t= 28.40s] done converged