A fly brain on a laptop

In 2024, researchers finished the complete wiring diagram, the connectome, of the adult fruit fly brain: 139,255 neurons and more than 50 million synaptic connections, every link mapped. The detail that stuck with me is what came next. Someone simulated that entire circuit, all 139,000-odd neurons and 50 million connections, and it ran on a laptop. Not a server farm. A laptop, the kind that overheats playing a video.

And it wasn’t a static diagram. The simulation predicted real behavior. It correctly identified which neurons fire when the fly tastes sugar, and which inhibitory neurons kick in when bitter is mixed with sweet to stop the fly eating something harmful. A wiring map, plus a simple model of how signals flow along it, reproduced functional responses of a living brain.

Two things make this matter beyond the novelty. First, the honest caveat: a connectome is not a living brain. It’s the road map, not the traffic. It doesn’t capture neurotransmitter chemistry, the influence of non-neuron cells, or any coupling to a body and an environment, so calling it a “simulated fly” oversells it.¹1 For scale, the human brain is around 86 billion neurons and roughly 100 trillion connections, six-plus orders of magnitude past the fly. That gap is not a detail.

Second, and this is the part that’s interesting from an AI angle, this is a real biological neural network that works, and architecturally it’s nothing like a large language model. It’s a different, evolved answer to the question “how do you build something that computes with neurons,” and at least one of the researchers floated it as an alternate path toward advanced AI. Whether or not that pans out, it’s a useful reminder that the LLM is one design point, not the design. Evolution shipped a working brain in a bug and didn’t use a single transformer.