Angle + power in, gravity does the rest — the same two lines of code flying the cannonball. Let students call out angles.
Same rule, new world: Orbit Lab 🪐
LIVE GAME
Identical gravity code — pointed at the sun instead of down — gives you the solar system. Flick too slow: crash. Too fast: escape. Just right: orbit.
A humbling truth · live
Chaos: tiny differences explode
Two double-pendulums, started 0.001° apart. Same rules, same computer — watch them disagree. This is why weather forecasts stop at ~10 days.
New idea · when the "thing" is everywhere
Smoke, heat, weather: chop the world into cells
🎈 A ball has ONE position. But smoke? Heat? Weather? They're everywhere at once
🔲 The trick: chop space into cells — each holds one number ("how much smoke here?")
🌬️ Advection: the wind carries it — chimney smoke drifting downstream
♨️ Diffusion: it spreads all by itself — perfume filling a room, milk swirling into tea
Every cell plays the same tiny game: look at your neighbors, update your number. That's ALL a weather supercomputer does — a billion times per second.
Computor 🧑💻 — YOU are one cell
LIVE GAME — YOU DRIVE
One volunteer drives; the room predicts. Every ▶ Step = one tick of the future — and cell #8's arithmetic appears in plain left / me / right words with the answer. Predict, step, check… then crank ⏱️ Δt past the red line and meet the famous 💥.
Draw it, then watch it flow: Wind Tunnel 🌬️
LIVE GAME
Doodle any shape and the wind peels off it into a Kármán vortex street — the same swirls behind cars, wings & flags. Real turbulence, simulated live.
Why it matters
Test it in the computer before you build it
✈️ Planes & rockets — thousands of crashes that never hurt anyone (my job!)
🌪️ Weather & climate — the whole planet in a supercomputer
🏥 Medicine — blood flow, drug design, surgery practice
🎬 Games & movies — every splash and explosion you've ever seen
Simulation meets AI: Train the Cannon 🎯🧠
LIVE GAME
Don't aim it yourself — collect practice shots in the simulator, train a model by gradient descent, and watch it learn to hit new targets.
The honest part
Simulation ≠ reality
Every simulation simplifies — perfect spheres, no wind, smooth tables
Small modeling errors + chaos = drift from the real world
Robots trained only in simulation stumble in reality (the "sim-to-real gap")
Rule of the pros: simulate, then TEST in the real world, then fix the simulator.
Recap
What you now know
🕰️ Simulation = state + rules + tiny time steps
⚖️ Step size is a trade-off: accuracy vs speed (too big → broken physics)
🌬️ Smoke & heat live on a grid of cells: advection carries, diffusion spreads — and steps past the CFL speed limit → 💥
🪐 One rule, many worlds: cannonballs and planets share an equation
🦋 Chaos limits prediction — even for perfect simulators
🤖 AI practices inside simulators — millions of free tries
Questions? 🙋
Up next at 9:45: AI World Models — what if the machine learns the rules itself?
Cannon, Orbit, Computor & Train-the-Cannon are all in the mini-games hub.