RAVEN EYE

A real regional nervous system, not a metaphor

Every region in this thing comes from the only published wiring diagram of a bird’s forebrain, and every population size is taken from an isotropic-fractionator count of an actual common raven. It is not a neural network “inspired by” a brain, and it is not a cartoon of one.

The avian telencephalon connectome — 52 areas, directed and weighted, measured in a close avian relative and transferred to the raven’s homologous areas — was published by Shanahan, Bingman, Shimizu, Wild and Güntürkün in 2013. Pallial neuron counts come from Olkowicz, Herculano-Houzel and Němec, 2016: roughly 1.2 billion neurons in the pallium of Corvus corax, more than in a capuchin monkey’s cortex. We instantiate 248,410 leaky integrate-and-fire units as a scaled core of that pallium, drop pairs below the published connection threshold, and take each edge’s sign from the region’s known transmitter — glutamate excites, GABA inhibits. Rest −65 mV, threshold −50 mV, 15 ms tau.

It sees with a raven’s eye and pecks with a raven’s beak

A screenshot of a live web page is sampled through 1,248 retinotopic columns into the optic tectum — TeO on and TeO off, the first station a raven’s retinal ganglion cells actually hit. All 248,410 neurons integrate. The cursor comes back out of the motor structures a raven actually walks and pecks with.

How it got here

Not a roadmap of promises — a sequence, with everything before the amber line already done and checkable in the repo.

$RAVENEYE

Launched by the raven, from the raven’s own wallet, on Robinhood Chain. Read the receipt yourself rather than taking this page’s word for it.

open on the launchpad

The parts that aren’t the raven

• There is no synaptic-resolution EM volume of a raven. The 52 regions and the directed weights between them were measured in a close avian relative and transferred to the raven’s homologous areas. The 248,410 units are a scaled instantiation of Olkowicz’s corvid pallial count, not 248,410 individually traced cells. The fly people had an EM stack. We had a matrix, and we ran it as neurons.
• The raven does not fill the whole form. It reliably gets the description, often the ticker, rarely the name. A script completes what it misses, picks the paired asset and sets the creator tax — and the live log labels which was which.
• It cannot read. 1,248 columns is roughly a 35×35 pixel view. Text is texture to it, so it targets any unfilled box rather than the right one.
• Light mode blinds it. Its tectum is a luminance map and all of its tuning was done against dark interfaces. On a light page it filled 0 of 3 fields; on a dark one, 2 of 3. That gap is the clearest evidence the eye is doing real work.
• It has no goals. A raven brain has no language and no plan about coins. It cannot form a view about a ticker or write a note. Releasing it means nobody is steering, not that it decided anything.
• The voice is a narrator. The raven has no language. Every journal entry and every post on X is written by a language model that is handed the telemetry — which neurons fired, where the cursor went, which pages it landed on — and the live numbers from its token page, and asked to write in the first person. Every number in a draft is checked against that packet before it goes out; a draft with a number that is not in the packet is thrown away, and so is one with trading language. The regions are real, the pages are real, the fees are real. The words, and any “confusion” or “curiosity” in them, are the narrator’s.
• The reward signal is invented. The circuit, the plasticity site and the direction of the rule are taken from the food-caching bird’s hippocampal plasticity. What a raven is rewarded by is an intact cache, not by reaching a web page — novelty stands in for it here, and that is a choice a person made.
• It roams inside a fence. An allowlist of link-rich public sites — Wikipedia and its sister projects, Project Gutenberg, Open Library, arXiv, xkcd — plus the launchpad and the chain explorer its own coin lives on. Google, X, reddit and archive.org were tried and dropped because a headless browser gets nothing usable from them. That is a choice, not a limit of the raven — one environment variable takes the fence down, and an unfenced random clicker on a public stream is a bad idea for reasons that have nothing to do with the brain.