deploying…
SCROLL · IT KEEPS CRAWLING
We took the complete wiring of a real worm's brain — all 302 neurons — and let it loose on Wikipedia.
Every page is a dish.
Every link is food.
It crawls. It backs away from what it can't eat, and follows whatever it swallows. Nobody steers it. Nobody trained it. The wiring is one hundred million years old and it still works.
Dataset: the full hermaphrodite connectome — White et al. 1986, as digitized by the OpenWorm project. The whole edge list is embedded in this page. View source: it's all there, neuron by neuron.
The network drawn above is not decoration. It is the actual edge list of the C. elegans nervous system — every neuron named, every connection weighted — embedded in this single HTML file.
The crawl is live. This page fetches real Wikipedia articles over Wikipedia's public API — open the network tab, the requests are right there. The article's own text is sniffed against a fixed scent lexicon (embedded in this file, inspectable), the result drives the real chemosensory neurons (AWC, AWA, ASE vs ASH, ADL), the signal spreads through the real synapses, and whichever command pathway is driven harder — forward (AVB, PVC) or reversal (AVA, AVD) — decides. When it swallows a page, the next dish is picked from that page's actual links. Nothing is scripted; the console shows every step. No connection? It replays a cached crawl and says so on the card.
One thing we did not invent: in the real wiring, the danger pathway (ASH → AVA) is roughly eight times stronger than the appetite pathway. A whiff of threat beats a feast. The worm is a coward, and it has survived everything for 100 million years by being one.
It is not an AI. It has no weights, no prompt, no plan.
It cannot be aligned, fine-tuned, or reasoned with.
It is a worm. And it is hungry.
$WORM
deploying…