139,255 neurons from a real fruit fly are about to feel a gravity no fly has ever felt. Every fly that ever lived knew 1 number, and it was 1g. This one is a copy, wired cell by cell from a real brain with 139,255 neurons and 50 million synapses, then dropped into a virtual body with 87 joints and real physics. The cage is an acoustic levitator with no floor, just sound holding it up. Here's the test: 1g → 2g, and every leg weighs double 0.2g, almost weightless then gravity rotates and down becomes up The brain is never told the number, and no line of code says "brace" or "flip". It gets 3 signals: acceleration, posture and leg contact. The same wiring once dodged a swatter in 100 milliseconds, and now it has to work out a new universe with fewer neurons than a small town has people. If it adapts, the next body is a robot. Nobody taught it gravity, and nobody is teaching it this one.

166,691 neurons from a dead fruit fly woke up inside the Matrix and learned a maze nobody taught it. The fly was cut into 66 pieces, 7 electron microscopes scanned them for a year, and AI stitched the slices back into 1 brain this month Cell published the map, every neuron and every wire. Then someone dropped that wiring into a 9x9 green maze with walls, a smell and a drop of sugar in the far corner. Trial 1: 61.7 seconds, 5 dead ends, 49 cells walked. Trial 12: 9.7 seconds, 0 dead ends, 27 cells. 6.4x faster, and nobody coded the route. Here's the loop, every 15 ms: Eyes and antennae turn walls and smell into 2.25M spikes a second. Steering neurons DNa01 and DNa02 turn the body while 6 legs walk in a tripod gait. Sugar hits and PAM dopamine cells fire: this way Wall hits and PPL1 fires: dead end. Deep in the mushroom body, the wires between Kenyon cells and output neurons shift, down 41% by trial 12. Nobody made that trick up for the demo real flies learn the same way: 1 sip of sugar paired with 1 smell, and they remember it for days. The run in this clip is an illustration, but the wiring is real. A fly brain is the size of a poppy seed, a mouse has roughly 420x more neurons, a human 516,000x more. Nobody taught the fly dopamine did.

33 years of human work went into mapping a brain smaller than a poppy seed. This is all of it, every one of its 139,255 neurons. 1 female fruit fly was sliced into 7,050 layers, each thinner than a virus, then shot in 21 million photos. AI traced the cells, and humans checked every wire by hand: scientists, students and gamers clicking for fun. The final map: 139,255 neurons 54.5M synapses 8,453 cell types 150 meters of wire folded into 1 speck No one had mapped a full adult brain before, not a mouse and not a human. A fly went first. In September 2026, Janelia and Google released the male, with 166,700 neurons mapped from brain to legs through an intact neck. 2 weeks later, the internet put the fly to work: It played Doom It flew a simulated drone with zero training It landed a simulated rocket 60 times out of 80 It wrote LinkedIn posts But the map only shows where the wires go, not how strong each one is. It has no voltages and no timing, like a circuit diagram with no values on it. Nobody knows what the fly is thinking, and it posts on LinkedIn anyway. It took 33 person-years to map 1 fly, and your head holds 617,000 of them.

An engineer built a turret that hits any pixel he clicks, and a fruit fly still dodges it in 100 milliseconds. 2 paper targets on a garage wall, blue air lines coiled around a geared aluminum turret, 1 click on the webcam feed and the green laser lands on the bullseye. A Python GUI talks to an ESP32 over a websocket, but every shot still starts with a human hand on a mouse. Then there's the target. Half of a fruit fly's brain is vision, and when something looms 2 cell types fire into a single giant fiber that spots the threat, plans the escape and sets all 6 legs in about 100 ms with no mouse, no websocket and no guy. Engineers are already wiring it into machines: one open-source drone runs a 138,639-neuron fly map with 0 training, steers away from anything looming and loses the dodge the moment you shuffle the synapses. Another team rebuilt the fly's eye as 45,669 neurons with 734 free parameters, where a standard AI net needs over 1M. The turret needs a camera, a GUI, a chip and a person, while the fly needs 1 mg of wiring. The turret needs a human to aim, the fly needs nobody to leave.

A dead fruit fly was sliced into 66 pieces. Now its 166,700 neurons are driving robots. Nobody trained it, no dataset, no reward function, no GPU farm, just the wiring. The fly took a second to die, but copying it took 7 microscopes running nonstop for a year, shaving it 1/10,000 of a hair at a time and photographing every layer before the next cut. AI rebuilt it from the scraps, eyes to legs with the neck still attached, 124M synapses and 11,710 cell types, 44 years of human labor finished by machines. Scientists first tried this in 2008 and guessed it would take 500 people 10 years. Then the internet got the file and someone gave it a body. Camera → fly eyes → 166,700 spiking neurons → descending neurons → motors. One of those neurons is DNp01, the giant fiber, the cell that fires when your hand comes down. Push an object at the camera and the looming detectors light up, so the robot swerves. It took 0 training runs, just the same reflex that gets a fly out from under a swatter in 100 milliseconds. Then came the control test, where they kept all 138,639 neurons and shuffled where the synapses connect. The behavior died. The intelligence was never in the neurons, it was in the map. The fly is gone, but the flinch is still running.

A pixel goldfish running Jev now thinks 193x faster than frontier LLMs, and somebody gave it a gun. It was supposed to be a cute desk pet on a matchbox-sized screen that blew bubbles and waited for food, but now it's wired by a coiled yellow cable into a vibe-coded arcade cabinet and it's hunting. Jev can't chat, write or explain itself, it only decides: give it the game state and up to 255 moves and it fires back in 70ms, while ChatGPT-class models need about 3 seconds, which is enough time to die twice in a corridor shooter. The fish narrates as it plays, "I'll take the—" and the shot lands before the sentence ends. Score 00070 → 00340 → 00560 → 01220 At the end a Pac-Man ghost floats into the crosshair and is gone before you notice it. An ex-OpenAI researcher raised $40M to build this and launched it this month, with input at $0.042 per million tokens and output free, so an AI that reacts faster than you costs less than fish food. Everyone is racing to build AI that thinks longer, and this one wins by barely thinking at all. The goldfish has a 3-second memory and a 70-millisecond trigger finger.

126 NEURONS NEED THIS WHOLE MONITOR WHILE A FRUIT FLY FITS 166,700 INTO A POPPY SEED, AND ANYONE CAN DOWNLOAD IT Every instinct here was typed in by hand: 21 inputs → HUNGER, ENERGY, SMELL ENEMY, SIGHT ENEMY 18 outputs → SPEED, BITE THRESHOLD, SOUND ALARM, HIDE 11 of those outputs still have no name, and the whole thing is under 3,000 wires packed into a solid red wall. The fly never needed labels. Its map came from 1 male fly cut into 66 pieces and shaved layer by layer under 7 electron microscopes for a full year. Within 5 days of release the internet had it "playing" Minecraft, Doom and Beat Saber, yet most of those demos barely use the eyes. Half the fly's brain, by volume, is built to see. In 2024 a Janelia and Tübingen team took the vision circuit alone: 45,669 neurons, 64 cell types, 1,513,231 connections. They froze the wiring and trained just 734 numbers instead of the 1 million+ a blank network would need, all for 1 task: tell which way the world is moving. The model never saw a real neuron and predicted them anyway, with T4 firing for light edges, T5 for dark, and 26 lab studies agreeing. The wiring did the thinking while training just turned the knobs, and it's all public: pip install flyvis A real fly spots a swatter, plans the jump and sets its legs in about 100 milliseconds. The red wall needs a neuron called HIDE the fly just leaves.

166,700 neurons from a dead fruit fly just walked across someone's kitchen counter. A 3D-printed robot crawls toward the camera on 4 jointed legs with sawtooth feet, orange servo wires strapped to its back with velcro. In the corner of the video, a brain glows. That's the fly. Scientists cut it into 66 pieces, ran 7 electron microscopes for a year, and let AI stitch the slices into 1 map: 166,700 neurons and 124 million connections, every wire from the eyes to the legs. One builder connected it to a Strandbeest. The simulated brain fires, the firing turns into motor commands, and the commands move the legs. A fruit fly walks on 6 legs and this one has 4, but the brain doesn't seem to mind. Every fly wants into the kitchen. This one had to die first.

A guy wired 166,700 neurons from a dead fruit fly into a creepy pair of robot eyeballs on his desk, and now they watch him. 2 plastic eyeballs and 2 blue servos sit on 1 Arduino, with orange wire running into a gaming keyboard that glows purple. The eyelids blink and the pupils drift left, then snap back, but nobody wrote that motion. The brain came from a real male fly, mapped down to every connection in its nervous system: 166,700 neurons and 124M synapses. To get there, scientists cut the fly into 66 pieces and shaved each one thinner than 1/10,000 of a human hair, while 7 electron microscopes ran for 1 year. By hand, the mapping would have taken 44 years of human work. The file is free to download, so he downloaded it. The circuits that once steered a fly away from a swatter now decide where 2 plastic eyes look. The fly's eyes were built from hundreds of tiny lenses and never had eyelids, and now it blinks. The fly died in a lab and woke up on a gaming desk.

Kagome Kagome was sung at a Tokyo expo in 2011 by a silicone mouth with 9 motors and no face. The room went quiet, because the sound was wrong in a way nobody could name, like a song being learned from inside a throat that had never owned one. A microphone sat bolted to the right side of the frame, and everyone assumed it was there to take orders. It was pointed at the mouth. The machine was listening to itself. Professor Hideyuki Sawada had given it nothing no recordings, no phoneme library, not one line of text, just an air pump for lungs, 8 vocal cords cut from silicone as soft as human mucous membrane, a tongue, and a resonance tube 180 mm long with a nasal cavity of 60 cm³ that opens for "m" and "n". So it babbled. 200 random sounds fed into a 25 × 25 map of 625 nodes, 6 people in the room saying yes or no, and that was the entire training set smaller than what a single Instagram ad gets tested on. It learned 5 Japanese vowels the way an 8-month-old does, make a noise, hear the noise, move the tongue 1 mm. The clip went around as horror, one headline asked if the robot apocalypse had started, and 15 years later it still gets reposted every few months with the same caption about nightmares. All of them missed who it was built for. 6 deaf teenagers came to that lab having never heard their own voices, only felt them in the bone, and they watched the silicone tongue take the shape of a vowel, copied it with their own, and 5 of 6 started speaking clearer. Every AI voice on your phone was trained on millions of hours of humans talking, and not one of them has lungs. This mouth has never read a word. It breathes, listens, and tries again.

AN 18-YEAR-OLD ENGINEER IN SHENZHEN WIRED 139,255 FLY NEURONS INTO A ROBOT THAT FORGES HIS OWN HANDWRITING The rig is 400 yuan of Taobao parts, 2 stepper motors and a 6 yuan gel pen clamped in a 3D-printed head. He fed it 40 pages from his own notebooks 11,000 characters, every loop and slant and every place the hand drifts at hour 14. Then he stacked 3 brains. The first writes the answers, chatGPT, plain. The second is his own 3 years of class notes in a vector store that rewrites those answers in his vocabulary, his sentence length, his errors, 2 wrong per page on purpose. The third is the one that shouldn't work. He loaded the FlyWire connectome, a complete map of a fruit fly's head: 139,255 neurons, 54.5 million synapses, sliced into 7,062 layers by a lab and published free, and he runs it as the motor controller, so the circuit that once timed a wing beat now times a pen. Every plotter on the market gets caught in 4 seconds because the strokes are too even machines write like machines, straight and spaced and dead. The fly can't hold a line, it shakes at 0.3mm, the same wobble as a tired wrist at 11 PM. That flaw is the whole product. 62 assignments in, 4 teachers, average mark 91, and one of them wrote in the margin: your handwriting has gotten more natural this term. The fly has been dead for years its wing still moves at midnight.

A 19-year-old at Stanford does 30 students homework, and his only employee is a dead fruit fly. The fly was real. Scientists cut its nervous system into 66 pieces, and 7 electron microscopes spent 1 year shaving and scanning them in layers 1/10,000 the width of a human hair. Google's AI stitched the scans back into a brain with 166,700 neurons and 124 million connections. Janelia spent 18 years chasing the largest brain map ever made, the journal Cell published it on September 3, and 2 weeks later it was doing high school algebra homework next to a brown couch. He feeds each student's handwriting in through the fly's eyes, pays it in dopamine for every stroke that matches and punishes anything too neat. The neurons that told 6 legs where to step now tell a blue pen where to go. It writes √5 in 30 different hands and has no idea what a square root is. The fly died for science and came back for homework.

150 drones, each run by a fruit fly's 166,700-neuron brain, living their own lives with 0 pilots. Nobody has built this yet, but every part of it already exists. First the map: Janelia and Google traced every neuron in a male fly's nervous system, 166,700 cells and 125M synapses packed into something smaller than a poppy seed. Then the pilot: coders wired fly brains into simulated drones, and one build flew with 0 training. Show it something rushing closer and it swerves, shuffle the synapses and the dodge is gone, because the reflex lives in the wiring. Then the swarm: another build copied 1 brain into 3 drones, and the README sums it up as "same wiring, separate spikes." Now scale that to 150. The fly already carries everything a drone needs: a ring of neurons that works as a compass in the dark, cells that fire when something in view grows too fast, and a way home, because after 1 sip of sugar a fly loops back to the spot by counting its own steps and turns. Swap the sugar for a charging pad. 150 drones lift off at dawn with no route and no mission, drifting, dodging birds and chasing light all day, until at 15% battery 150 compasses swing the same way and they land in rows, charge and leave again. They wouldn't need us, only our outlets.

166,700 neurons from a male fruit fly's brain are teaching a pair of silicone lips to talk. A steel frame, a silicone tongue, a row of motors under the jaw and a microphone pointed straight at the mouth, with no speaker anywhere. Air pushes through artificial vocal cords, the motors bend the lips and the microphone catches every sound it makes. If the pitch is off, the motors shift before the next try. Nobody coded the words: it finds each lip shape by trial and error and uses old sounds to guess new ones. Babies learn to talk the same way: babble, listen, fix, repeat. The new part is the brain in the corner of the screen, with 166,700 neurons and 125 million synapses packed into something smaller than a poppy seed. Scientists spent 10 years mapping every wire, and the job took 44 years of human labor. Male fruit flies already sing: they buzz 1 wing to court a female, and she hears the song through her antennae. That song never had a mouth until now. A trained linear readout turns its spikes into words and the mouth does the rest. You needed 86 billion neurons and 12 months to say "mama." The fly needs 166,700 and a microphone.

166,000 neurons from a dead fruit fly now pilot a robot that drives and flies. It's the wildest piece of engineering on the internet right now, and the pilot has been dead the whole time. The robot rolls across a bedroom floor like a toy until the wheels stop being wheels: 4 propellers inside the rims spin up, and it lifts off, hangs at desk height, turns and lands. The body came out of a cardboard box of drone parts, and the brain it copies is smaller than a poppy seed and was never designed by a human. Janelia and Google spent years turning 1 male fruit fly into a spreadsheet, with every neuron and up to 125M synapses labeled, and on September 3 they posted it free for anyone to download. To build it, the fly was cut into thousands of layers and traced cell by cell under an electron microscope. Camera frames get turned into what a fly's eyes would see. The spikes run through the real wiring, where each neuron is 1 equation that rests at -52 mV and fires at -45. When something rushes at the lens, LPLC2 and LC4 light up, the same cells a fly uses to dodge your swatter, while DNg02 at the far end sets climb and turn. Inside the brain there's no flight code, no PID loop and no line that says "avoid the bed." Tesla's self-driving learned from billions of miles of driving, and this pilot learned from 400 million years of flies that didn't get swatted. A real fly walks on 6 legs and flies on 2 wings, and this one got 4 wheels and 4 propellers. The fly is dead, the brain is a file, and the file just took off.

A fruit fly just carried a 118 g banana home, 147,500x its own body mass. The banana is fake, but the brain steering it is not. 4 neurons in there do something ChatGPT still can't. A fly leaves a drop of food and wanders off in the dark, 1.45 m of loops in the clip, then beelines back 77 cm dead straight. To do that, its brain keeps a running sum of every turn, called path integration. The neurons that report each step are known, but the neurons that add them up were never found. Most models say the sum lives in firing: neurons excite each other in a loop so tight the signal never fades or blows up, the same way RNNs and LLMs hold state. Then a researcher ran the full male fly connectome and found the loops are thin, with self-connections in the hΔ cells carrying only 1-2% of input. So maybe nothing keeps firing, and each step gets written into synapse strength until the weights become the map. 4 neuron types have every part that plan needs: hΔH, hΔA, hΔI, hΔG. Step signals in, with 1,010 synapses onto hΔH alone Speed-tuned dopamine to gate each write Octopamine, the reward signal, to wipe the slate at the food In simulation, an agent on this store gets back within 1.1 units of the food, versus 6.8 without it. No one has recorded hΔH in a live fly yet, but 3 tests are already written up. If it holds, this is fast-weight continual learning in a brain smaller than a poppy seed. ChatGPT stopped learning the day it shipped, but the fly learns with every step.

166,000 neurons from a dead fruit fly now hunt drones from a bedroom desk. A white drone the size of a palm lifts off the carpet, and a black 3D-printed turret on the desk swings after it, with the fly's brain working its 2 stepper motors, 1 camera and 2 barrels through a nest of rainbow wires on a breadboard. That brain sits in the top-left corner of the screen and flares every time the drone jerks. A Patriot interceptor costs about $4 million, and armies are firing them at drones worth a fraction of that. This brain is smaller than a poppy seed, and Google put it online for free 2 weeks ago. Then someone gave it a job it was built for. Most of a fly's brain exists to see motion, and this brain is male, and male flies spend their lives chasing whatever moves. The drone climbs past the ceiling fan and the crosshair climbs with it, then drops when it drops. For 18 seconds it never loses the drone. In the last shot the drone flies home to the guy holding the controller, and the crosshair follows it right onto his chest. It spent its whole life dodging swatters. Now it's the one aiming.

166,700 neurons from a fruit fly are now walking a machine across my floor. Scientists rebuilt the map every neuron, 25.6 million synapses, traced out of one Drosophila and released to anyone who wanted it. I took it out of the computer and put it in a body. A Strandbeest PVC pipe, 12 legs, no motor in the original design. The panel in the top-right is the brain running live, and every spike it fires gets translated into a motor command that moves the legs below. Nothing was trained here. No reward function, no dataset, no gradient descent the wiring was already finished, I only built the interface between its outputs and the machine. It's a computational model, not a resurrection, and it doesn't know it has legs. It fires the way it fired inside the skull, and the pipe moves. The fly never learned to walk like this and it's walking anyway.

A Stanford student put 166,700 fruit fly neurons in a robot that now writes homework for 30 students. Not one of them has picked up a pen since. The robot is a black aluminum rig with a blue ballpoint, writing square roots and fractions line after line in handwriting that doesn't look like a machine's, while the fly's neurons glow in the corner of the screen. Each student gives it a handwriting sample, and the fly learns the slant, the spacing, the size of the loops and where the letters get lazy at the end of a line, then writes as that student. 30 notebooks, 30 different hands and 1 brain. That brain is smaller than a poppy seed, and a human brain has over 500,000 times more neurons, but the fly doesn't get tired, doesn't smudge the ink and never mixes up whose 7s have a line through them. The teacher checks 30 notebooks, and a fly wrote every one.