I built a very small language model on plebslop. How small? Small enough to fit in a single *standard* bitcoin transaction. There is now a plebslop generator on every Bitcoin node. The UI includes client-side inference code. It loads the weights and does it all in-browser