Log #1. A random walk, then 1,000 bots that steer themselves.
october 2026 • jax, numpy, swarms • 1 min read
1,000 dots, each taking random steps. Nothing smart yet. The point was the plumbing: write the same thing in NumPy, then in JAX, and learn why JAX wants pure functions and explicit random keys. That's what lets every bot's logic run in parallel later.
1,000 bots, each looking only at neighbours within a small radius, follow three rules: don't crowd, face the way your neighbours face, and drift toward them. Nobody leads, yet the whole group moves as one.
The interesting part was speed. Checking every bot against every other bot is a million comparisons a step. Splitting the space into a grid of cells, so each bot only checks the cells around it, is what makes 1,000 bots run smoothly, and it's the same idea every later phase relies on.
What I took from it: "no central brain" isn't a slogan, it's a constraint you can check. If any rule reads global information, it's cheating.