(@@) Immortal Fruit FlySourceGet $FLY

The fly-brain trend

How a connectome paper became a meme in nine days, and what everyone actually built.

Timeline

DateEvent
Oct 2024Shiu et al., Nature: whole-brain leaky-integrate-and-fire model of the FlyWire connectome. The recipe everyone copies.
8 Jun 2026MaleCNS v1.0 public (Janelia + Google): 166,700 neurons, whole male central nervous system, CC-BY.
3 Sep 2026Cell paper and Google Research blog post.
4–6 SepMinecraft, DOOM, Fabric mod, Bad Apple on the fly's brain.
9 SepBeat Saber, ~22M views.
10–11 Sep$FLYBRAIN launches on Robinhood Chain, peaks ~$54M after Marc Andreessen follows; $HER, bFlyBrain (BSC) follow.
13 SepThe Strandbeest: “the fly brain escaped the computer.”
15 Sep 04:07Siyuan (YZi Labs): on-chain / immortal fruit flies on BNB Chain.
15 Sep 04:15CZ: “Would be cool to see someone make ‘immortal fruit flies’ on BNB Chain.”
15 SepImmortal Fruit Fly genesis brain deployed to BSC mainnet.

Projects

ProjectWhoWhat
Eon Systems embodied flyShiu, Harris et al.NeuroMechFly body; grooming, feeding emerge; 15 ms brain–body sync
NeuroCraft (Minecraft)@evnsnclrfull 166,700-neuron MaleCNS, first viral post (4 Sep)
DOOMFLY@nftechie_ViZDoom; 3,335 R1–R6 + 811 R8 inputs; DNp20 turning; PPL1 punishment
Fly Brain Minecraft mod@BlendiBylFabric mod, Shiu LIF, 176k neurons, per-mob brains
Fly64ornataSuper Mario 64; DNg100 forward, DNa02/DNg13 steer, DNp01 jump
Fly Dinocobanov80-neuron circuit, 99/100 courses after CEM training
Beat Saber@_lyraaaa_~22M views; motor track trained on a replay
flyversedjmangoRust LIF, 6,091 photoreceptors, 0.23× real time
flypokevshapenkopoke a neuron in the full FlyWire brain on a laptop
DesktopFly / gnatDenisSergeevitch / lubabs770668-neuron escape+walk circuit as a desktop pet
Strandbeest robot@Frank_web33“The fly brain escaped the computer”; 14k likes
Overwatch / parallel parking / Bad Apple@derpchud / @alright_mark / @linguinelabsthe meme wave
$FLYBRAIN, $HER, bFlyBrain, Richyvarioustokens; brain runs off-chain, only swaps are on-chain

The common recipe

  1. Download the 1.1 GB MaleCNS weights file (or FlyWire v783), keep the retained neurons, sign edges by predicted neurotransmitter.
  2. Shiu-style LIF: rest −52 mV, threshold −45 mV, 0.275 mV per synapse, τ 20 ms, 0.1 ms step.
  3. Render the game, sample pixels into photoreceptors as Poisson spikes.
  4. Read a handful of named descending neurons (DNa02, DNp09, MDN, DNp01, DNp20) as a joystick, with a hand-made or trained readout.
  5. Optionally pulse dopamine neurons (PAM11 reward, PPL101 punishment) to depress KC→MBON synapses. No project has demonstrated task learning from this.

What none of them did: put the neurons on the chain. See the vision.

Full curated list: awesome-fly.