A grid of tiny magnets, each nudged by its neighbors and jostled by heat. Drag the temperature and watch randomness turn into structure — the way a computer can compute with noise.
Each cell is a spin pointing up (■) or down (■). Every step, a random spin
considers flipping: if flipping lowers the energy (agrees with more neighbors) it usually does; if it raises the energy,
it still might — with a probability set by the temperature. That's the whole rule. Hot = the jostling wins, everything is noise. Cold = agreement wins, order crystallizes.
In between, at the critical temperature Tc ≈ 2.27, the two are balanced and the pattern seethes at every scale at once.
This is not a screensaver. It's the substrate of probabilistic computing: give the spins a pattern of couplings that encodes a problem, and a low-energy
state is a good answer. But here's the catch this toy makes visible — if you just cool it fast (drag straight to cold), it freezes into a stuck patchwork, a decent-but-not-best local minimum
it can't escape. Reaching the global best is genuinely hard. That's the entire reason for the "Slowly cool" button: annealing — easing the temperature down so the noise can
still jiggle the system out of bad states on the way — is how you coax it toward the real answer. The noise isn't the enemy of the computation. It's the only thing that lets it escape being wrong.