zebralink fish
A real zebrafish's brain watches its own token and decides when to hunt.
The subject
The fish is a larval zebrafish, Danio rerio, about a week old and four millimetres long. Its whole brain was filmed cell by cell under a light-sheet microscope: 71,721 neurons over 7,879 frames, released as ZAPBench by Google Research and HHMI Janelia. From that recording we picked 545 cells that answered to light, turning, reward and threat, and fitted how each one follows the others. The fit is to its neurons, never to prices. Nothing is trained on the market, and no AI model takes part in its decisions.
What it senses
Buys drift past as prey. Every trade in its pool becomes small bright specks in the water, like the paramecia a real larva hunts. The more traded, the more prey.
A falling price looms as a shadow. A dark disc grows over its visual field, the stimulus that sends a real larva bolting.
A new high reaches its reward cells. The water brightens when the price climbs and dims when it slides.
No trades is still water. Nothing moves, and its cells settle.
What it does
Hunts. When its hunting cells cross the line, it strikes at the prey. Each strike is its buy-back-and-burn signal.
Escapes. When the shadow fires its escape circuit, it bolts. A frightened fish does not hunt.
Rests. Below both lines it hovers, the way a larva idles between hunts.
Its water
The fish lives in one pool: Zebra / GOOGL on Robinhood Chain, launched on Pons and paired with Google's stock token, because the brain behind it was recorded by Google Research. Its first hours were a feeding frenzy. Then a shadow took almost all of the price in one afternoon. The water has been quiet since.
Watch it
The body view shows the fish at its desk, acting out its latest decision. Through its eyes shows the exact picture its light cells read, with every one of its 256 light cells marked. The brain view shows all 71,721 neurons with the 545 that decide lit up. Status, Senses and Activity show the numbers behind each choice. Every viewer sees the same journal, computed from the same market record.
Science
Recording: ZAPBench, Lueckmann et al., Google Research & HHMI Janelia, 2025. Prey capture after Bianco, Kampff & Engert, Front. Syst. Neurosci. 2011. Looming escape after Temizer et al., Curr. Biol. 2015. The Mauthner escape circuit after Korn & Faber, Neuron 2005. Model: ridge AR(3) over 545 cells, median R² 0.94 on held-out frames.
