Over the last week, the internet learned that you can download a fruit fly connectome. It reacted as it usually does: putting a fly brain in Minecraft, playing Bad Apple on it, trying to make it play Doom, and using one as a Super Mario 64 controller. “Flyslop” is a pretty good name for the resulting genre, and we’re loving them!
The immediate trigger was the September release of the MaleCNS connectome: a fully proofread map of the adult male Drosophila central nervous system spanning brain and nerve cord, with about 166,700 neurons. This was not the first complete adult fly brain (the FlyWire adult female brain arrived in 2024), but the intact brain-to-nerve-cord map makes sensory-to-motor experiments much easier.
A few particularly good specimens
- Minecraft fly: NeuroCraft Fly runs a retained MaleCNS graph in Minecraft, then uses hand-chosen neural readouts to modulate scripted body programs.
- Bad Apple, but on a fly brain.
- DOOMFLY: Doom frames stimulate modeled sensory neurons; activity propagates through the MaleCNS graph; selected neurons are mapped back to game controls.
- Fly64: A Mario 64 hack. The author describes it as 100% vibe-coded and not reviewed for anything “real,” which is almost the perfect flyslop disclaimer.
- Fly “emotion recognition”.
And more…
For the first time, scientists have mapped the complete brain and central nervous system of an adult male fruit fly — a key model organism in science. 🪰 Working alongside HHMI Janelia Research Campus and the scientific community, @GoogleResearch scientists and researchers used Show more
the fly is deploying databases
i trained the fly to solve rubiks cube is there anything this little guy can't do?
in all my years of doing neuroscience, I never thought we would end up here
trained a fly to fly a plane
I turned the fly bisexual Inspired by Kallman et al ('15), I blocked mAL output in a 166,606-neuron fly model and measured responses in 8 candidate P1 male courtship neurons. In 3 experiments, when cued by other males, baseline P1 spikes rose from 0 to 1–5 after mAL blockade
I wireheaded the fly and forced it to doomscroll flytok. Dopamine neurons are measured and artificially enhanced to ensure maximum enjoyment. My goal is to create a fly that is happier than all other flies combined.
the fly is using the blinkers 😭
i trained the fly to cut doner kabab with a knife yours solves rubiks cubes, mine cuts doner. guess which one has a job in this economy
fly brain has learned python
Meyve sineği beynini TCMB’ye bağladım ve faiz kararını belirlemesini istedim. Sinek, insana göre genelde daha düşük faiz kararı belirledi ve enflasyon oranı benzer çıktı.
I trained the fly to play Super Mario
But can it run DOOM?! Well...actually, yes.
The scientific simulations are real
What makes all of this interesting is that there is now serious evidence that a connectome can be used to build models and make accurate scientific predictions. A beautiful example is Lappalainen et al. (Nature, 2024). They built a recurrent neural network based on the measured Drosophila visual connectome. The model covered 64 cell types in the motion pathway. Instead of fitting every neuron to recordings, they optimized a small set of unknown neuronal and synaptic parameters so that the network performed a biologically relevant computation: estimating visual motion.
The resulting model predicted neural responses across the visual system and agreed with measurements collected across many previous experimental studies. This indicated that anatomy plus a task was enough to recover substantial physiology that hadn’t been directly fit neuron by neuron.
A second line of work has similar results. Shiu et al. (Nature, 2024) constructed a leaky-integrate-and-fire model of the entire adult fly central brain using the connectome and predicted neurotransmitter identity. It is a deliberately simple neuron model, but perturbing it produced useful predictions about feeding and grooming circuits, some of which the authors then tested experimentally. What this showed is that the whole-brain wiring diagram contains enough structure to make nontrivial, falsifiable predictions. These results, that a connectome is very valuable for understanding and predicting neural activity, have been shown in many other Drosophila papers.
In March, we took the Shiu-style whole-brain model, combined it with the published NeuroMechFly body in MuJoCo, fed virtual sensory events into identified pathways, and used selected descending neuron activity to drive body-level controllers (but specifically not individual motor neurons). The fly could forage, groom, feed, and respond to inputs in a closed sensorimotor loop. We published a technical description and limitations because those limitations matter at least as much as the demo.
How to validate connectome simulations
A connectome is an extraordinary measurement, but it is not a frozen executable brain. It tells us which neurons are connected and, depending on the dataset, gives synapse counts, cell types, neurotransmitter predictions, and anatomy. It does not directly measure the moment-to-moment physiological state of the living animal: membrane potentials, channel states, neuromodulators, peptide signaling, hormone levels, short-term synaptic dynamics, and many forms of plasticity. Synapse count is useful, but it is not the same thing as physiological synaptic strength.
The interfaces are also engineered. A game has pixels and buttons; a fly has ommatidia, sensory neurons, descending neurons, a ventral nerve cord, muscles, and a body. Someone has to decide how game pixels stimulate the modeled sensory system and how modeled neural activity becomes “move left,” “jump,” or “fire.” Those choices can be biologically informed, learned, or completely arbitrary.
The right standard is validation: does the simulated nervous system reproduce neural activity and behavior that we can measure in the real animal, especially for interventions it was not tuned on? Lappalainen et al. and Shiu et al. are interesting precisely because they move in that direction.
What Eon is working on
At Eon, we are working toward whole-brain emulation (“uploading”) where the target is a particular biological brain, captured in enough structural and functional detail that a simulation can reproduce the relevant internal dynamics and especially the behavior of the original.
That means pushing well beyond today’s fly demos: collecting synaptic-resolution structure at much larger scale; measuring neural activity and structural connectivity in the same cells so that the mapping from anatomy to dynamics can be predicted at higher accuracy; building better neuron and synapse models; and validating the resulting emulations against held-out recordings and perturbations. The fly is useful because it gives us the first system where the entire stack can be run end-to-end.
Eon is working to scale brain emulation technology to upload the first humans. We’re building the tech stack to be able to collect larger connectomes and simulate them, starting with the mouse. Faithful whole-brain emulations will require a massive amount of data collection, model refinement, and compute. If you’re excited about this, contact us, or apply to our open positions!
AI is moving quickly
But the good news for whole-brain emulation is that AI makes the entire pipeline easier. The new male fly connectome is itself an AI story. The Google Research/Janelia effort used automated reconstruction to turn enormous electron-microscopy datasets into neuron shapes and connectivity. The 2024 Lappalainen work used deep-learning optimization to infer plausible neural dynamics from wiring plus task constraints. In connectomics generally, segmentation, tracing, proofreading assistance, model fitting, and code generation are all getting dramatically faster.
Of course, the amusing and fun evidence is the flyslop itself. A major connectome was released, and within days people had built Minecraft, Doom, Mario, and music-video experiments around it. Most of those are not traditional scientific results. But they demonstrate something important: connectomes are becoming programmable objects. You can download one, simulate it, connect it to an environment, test a hypothesis, and iterate.
While the core driver of flyslop is the amazing power of AI, the world is also recognizing the significant risks of “unaligned” black-box AI. Frontier labs are struggling to contain the AI agents they themselves built. At Eon, we believe that brain emulation is a potentially safer path towards digital minds that share human values, because they are human.
Collecting connectomes is getting exponentially cheaper. Connectome models will get better. The memes will get stranger. And sometime soon we expect a new experimental science of making brains executable.
Links & references
- Berg et al. (2026), “Sexual dimorphism in the complete Drosophila male central nervous system connectome,” Cell.
- Google Research (2026), “A connectomics milestone: Mapping the complete male fruit fly brain.”
- Dorkenwald et al. (2024), “Neuronal wiring diagram of an adult brain,” Nature.
- Lappalainen et al. (2024), “Connectome-constrained networks predict neural activity across the fly visual system,” Nature.
- Shiu et al. (2024), “A Drosophila computational brain model reveals sensorimotor processing,” Nature.
- Eon (2026), “How the Eon Team Produced a Virtual Embodied Fly.”
- NeuroCraft Fly (Minecraft) project page.
- DOOMFLY project and validation notes.
- Fly64 / Super Mario 64 project.
- Bad Apple fly-brain post (@linguinelabs).