They mapped a fly’s brain. The internet did the rest

Erick Sergio Valenzuela AguilarCo-Founder & Head of Product

September 13, 2026 · 4 min read

CoverOne representative neuron from each of the dataset’s 11,710 types. MaleCNS · FlyEM (HHMI Janelia), University of Cambridge, MRC LMB and Google Research · CC BY 4.0

This week brought three stories that look unrelated but together describe the technological moment we are living in quite well: we can reconstruct a brain, put thousands of AI agents to work on a mathematical problem open for almost a century and, at the same time, our teenagers are posting some of the worst educational results in decades. This is the first one, without the detours.

What MaleCNS is

Researchers from FlyEM (HHMI Janelia), the University of Cambridge, the MRC Laboratory of Molecular Biology and Google Research reconstructed the complete central nervous system of a male fruit fly (Drosophila melanogaster), brain and ventral nerve cord, at synaptic resolution. The dataset was released as MaleCNS v1.0 on 8 June 2026, and the paper describing it came out in Cell on 3 September.

• 166,700 neurons.
• 11,710 neuron types.
• Around 125 million synaptic contacts.

So far, serious neuroscience. Then the internet arrived.

Beat Saber and a bitcoin account

Within days, developers began running simulations of the connectome, using leaky integrate-and-fire neuron models, and wiring its sensory and motor neurons into digital environments. The video that became the reference point was posted by @_lyraaaa_ on 8 September: a MaleCNS-based simulation “playing” Beat Saber. It has passed 22 million views.

The quotation marks around “playing” are deliberate. The author later clarified that the video does not yet show the model playing on its own: its motor system had been trained to reproduce a recorded movement sequence, while seeing the blocks and reacting to them was still being developed with reinforcement learning.

Two days later came Stonkfly. Software engineer Alex Wormuth gave a simulation of the same connectome $100 to trade bitcoin: dopamine neurons are stimulated when the portfolio gains, and neural activity decides whether to buy or sell.

Its documentation is among the most honest writing on the subject. The fly does not see prices: it sees the pixels of a 320 × 180 chart projected onto its photoreceptors. “Buy” and “sell” come from comparing the activity of two descending neurons, one on each side, under a rule the author calls “an engineered interface, not a discovery of buy neurons”. And it says so plainly: the repository’s tests have not demonstrated profitable learning or strategy improvement.

The important caveat

Nobody uploaded a fly’s consciousness to a computer.

The connectome describes the wiring: which neuron touches which, and how many times. It does not include electrical activity or how each synapse responds. Turning that map into something that moves requires approximating each neuron’s activity mathematically, and deciding by hand where sensory information comes in and how activity becomes movement.

For the first time, a biological neural architecture at this scale is available for computational experiments.

And that is exactly what is fascinating. The viral videos are the least of it: what changed is that anyone with a computer can download an animal’s complete nervous system, under an open licence, and start asking it questions.