At attacher can guess, and has a 1/16 probability to guess right, but they have only one chance to guess because if you guess wrong, the process terminates (if it's a user-process) or the kernel panics (if it's in the kernel), so in the next opportunity you'll have it will be a different tag to guess.
One obvious optimization that would half the size of the data and also solve the gaps problem is to keep every border between two states once and not twice (for each polygon).
This would require some processing of the geometry to find the intersection points, but assuming that in the original data, the border between two states is the same exact line, it shouldn't be hard.
There was once a similar page that generated a game-of-life game with randomized rules. Does anyone has that link?
The concept there was more restricted, but that made it more likely to produce interesting results.
Though these two guys could not be more different from each other. Ivan actually has a toy marble machine video which he seemingly build over the course of a few weeks. I feel like he could design and build from scratch a fully featured MMX in not much more than that.
Feels like this article was cut in the middle just before the interesting part, where he's supposed to explain why welding for construction was not automated a much and how all this is relevant to AI...