Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

the fly can play deadlock I've successfully trained the MaleCNS v1.0 fruit fly connectdome's 166,700 neurons using YOLOv5 and video pretraining data to play deadlock in a keyboard environment with attention-based aiming. The fly has found the most success on Graves. #deadlock

wormy

1,709 subscribers

2,597,854 views • 4 days ago •via X (Twitter)

Education Gaming Science & Technology #deadlock

Anya Rossi• Live Now

Private livecam show

35 Comments

Unfazed_7994 days ago

“The fly has found the most success on Graves.” Greatest slander of all time

Orche4 days ago

It's Graves. Of course a Fly can play it.

assuka4 days ago

i need to make something like this but a fly watches your screen and is trained to turn on your vibe when it thinks it should

wormy4 days ago

fly x lovelock coming soon

assuka4 days ago

EXACTLYYY, my lovelock fork is up and i have my own version now since the og dev kinda removed the repo: deadass

Necromancy884 days ago

I'm a Graves player. You've given her haters unlimited slander ammo. Thanks a lot bro.

wormy4 days ago

Video PreTraining (VPT) allows the RL agent (the fly) to perform training on unlabeled video data and analyze expected supervisor values during an episode. I had the fly watch my ranked teammates but somehow he plays better than them. More here:

wormy4 days ago

the fly initially tried playing McGinnis but he got too bored.

Paxaral4 days ago

I genuinely can’t tell if this is a shitpost or not. But if not, this is awesome

Kira4 days ago

watch the fly hitting eternus with warden golden egg

~Raz. ♤4 days ago

oh my god the slander that will come out of a fly preferring to play graves 😭

🎩🐳🍷4 days ago

Put the fly on Macrill and watch it win every game

Slush4 days ago

i parry that heavy melee everyday of the week #flyswatter

Die Chance4 days ago

What rank is the fly

ClearVis🏳️‍⚧️4 days ago

Put it in ranked and see how high she climbs

c a t a t o n i c.4 days ago

@moragopus LMAOOOOO

ZeroMeOut4 days ago

Can the fly hit eternus tho?

QoR GalAxy 👺4 days ago

@moragopus Brother you are goated

NkoKirkto‽4 days ago

Graves is so easy and boring a fly can play her

Cherry4 days ago

@thnksfrthmsryOW every day new slander for a character i play comes out and i just have to fuck it we ball

Vita🌸4 days ago

I love this so much! How do you know it pfefers Graves? Any idea why? I wonder if her weapon having the auto-aim attach mechanic generates lot of positive feedback on hits, thus the fly believes it is performing better than on other picks where aiming is more demanding.

Wevo4 days ago

"The fly has found the most success on Graves"

Erucardu4 days ago

Top tier stuff I'm crying 😂😂😂

RayBuol4 days ago

NO what the fuckkkkkk nooooooohohohohoohooo

HuffFluff4 days ago

incredible slander

Devo 4204 days ago

Any other heros the fly likes? This shit is so crazy to me

Rio4 days ago

Is it playing against bots or players?

Cravvnn4 days ago

Lmfaooo

Some_Huge_Stuff4 days ago

Real talk tho this is some SoulKiller type shit, this fly is now living thousands of lives, some good, some bad, but never ones that are real Switching forms and body’s, knowing things it never knew, and being broken down into basic components This fly is Johnny silver-hand

Gelidic Havoc!4 days ago

Fuckkk you didnt have to doxx me like this fuckkkkkkkkk

Badass Yuki4 days ago

Even a fly can play deadlock right, but my team not

Some_Huge_Stuff4 days ago

Ok so this officially proves that it takes no skill to play Graves

Holderiano4 days ago

Amazing graves slander

HotPockat4 days ago

Warden mains sighing in relief that the character a fly can play the best is not theirs

Arua4 days ago

@juhhlea mosca jogando de graves

Related Videos

the fly can play guilty gear I've successfully trained the MaleCNS v1.0 fruit fly connectdome's 166,700 neurons using YOLOv5 and video pretraining data to play gear in a keyboard environment with attention-based neutral The fly has found the most success on Nagoriyuki. #GGST

the fly can play guilty gear I've successfully trained the MaleCNS v1.0 fruit fly connectdome's 166,700 neurons using YOLOv5 and video pretraining data to play gear in a keyboard environment with attention-based neutral The fly has found the most success on Nagoriyuki. #GGST

rin 🏳️‍⚧️

569,774 views • 3 days ago

I gave a simulated fruit fly brain access to the market. 166,700 neurons. Over 25 million connections. It trades new pairs using paper funds.

I gave a simulated fruit fly brain access to the market. 166,700 neurons. Over 25 million connections. It trades new pairs using paper funds.

CYBER_BULLY

53,034 views • 5 days ago

Ludwig says the vast majority of players have given up on CS2 and moved to Deadlock instead “I’ve gotten so many messages to play Deadlock, I’ve realized most of the CS2 community has given up on Counter-Strike and started playing Deadlock.”

Ludwig says the vast majority of players have given up on CS2 and moved to Deadlock instead “I’ve gotten so many messages to play Deadlock, I’ve realized most of the CS2 community has given up on Counter-Strike and started playing Deadlock.”

veizau

99,913 views • 18 days ago

Simulating a fruit fly's neural network using custom WebGPU kernels, running locally in a browser. The demo uses real connectivity data for 166,700 neurons from the MaleCNS connectome. Paint neurons and watch their activity drive the fly's movements. Built with GPT-6 Astra.

Simulating a fruit fly's neural network using custom WebGPU kernels, running locally in a browser. The demo uses real connectivity data for 166,700 neurons from the MaleCNS connectome. Paint neurons and watch their activity drive the fly's movements. Built with GPT-6 Astra.

Xenova

156,144 views • 8 days ago

I gave the fruit fly brain a fruit fly body and put it into a fruit fly sim. Now you can watch him eat, fly, and shit live on

I gave the fruit fly brain a fruit fly body and put it into a fruit fly sim. Now you can watch him eat, fly, and shit live on

Misch Strotz

41,609 views • 4 days ago

Introducing FlyOCR 🪰 - I trained a fly brain to read a PDF It uses the full MaleCNS v1.0 fruit fly connectcome. The architecture is inspired by doomfly by The fly splits a pdf image into individual glyphs, maps pixels into receptor activations, runs simplified current-based dynamics across the 166k neurons and 25m edges in the circuit, applies a compact readout model on the downstream spikes, and concatenates everything into the parsed output. On reading an actual Microsoft 10-k, the fly gets ~86% over the balance sheet heading, but is largely able to read the numeric values correctly. Over 1.7k+ sampled glyphs (chars+digits) it gets 87% accuracy. With enough training it might match some of the latter-generation MNIST models! Maybe eventually we’ll replace our doc parsing VLMs with flies. Full video below. Repo with full code + report:

Introducing FlyOCR 🪰 - I trained a fly brain to read a PDF It uses the full MaleCNS v1.0 fruit fly connectcome. The architecture is inspired by doomfly by The fly splits a pdf image into individual glyphs, maps pixels into receptor activations, runs simplified current-based dynamics across the 166k neurons and 25m edges in the circuit, applies a compact readout model on the downstream spikes, and concatenates everything into the parsed output. On reading an actual Microsoft 10-k, the fly gets ~86% over the balance sheet heading, but is largely able to read the numeric values correctly. Over 1.7k+ sampled glyphs (chars+digits) it gets 87% accuracy. With enough training it might match some of the latter-generation MNIST models! Maybe eventually we’ll replace our doc parsing VLMs with flies. Full video below. Repo with full code + report:

Jerry Liu

36,481 views • 2 days ago

I and nico ran the fly connectome and found the group of neurons (hΔH, hΔA, hΔI and hΔG) that could allow the fly to navigate using fast synaptic weight updates, not neural activations. This is fast-weight continual learning in a fly, something current LLMs don't do!

I and nico ran the fly connectome and found the group of neurons (hΔH, hΔA, hΔI and hΔG) that could allow the fly to navigate using fast synaptic weight updates, not neural activations. This is fast-weight continual learning in a fly, something current LLMs don't do!

Peter Wang

589,418 views • 1 day ago

THE FLY BRAIN ESCAPED THE COMPUTER. I PUT IT INSIDE A PHYSICAL ROBOT. Scientists released a reconstructed map containing 166,700 neurons from a fruit fly. I connected it to a Strandbeest. The panel in the top-right shows the simulated brain running. Its neural activity is translated into motor commands that move the mechanical legs below. The catch: this is a computational model built from biological wiring. I engineered the interface connecting its outputs to the robot. First, the fly brain entered DOOM. Then it played Chrome Dino. Now it has a physical body. 166,700 neurons. 25 million connections. One nervous system walking through the real world.

THE FLY BRAIN ESCAPED THE COMPUTER. I PUT IT INSIDE A PHYSICAL ROBOT. Scientists released a reconstructed map containing 166,700 neurons from a fruit fly. I connected it to a Strandbeest. The panel in the top-right shows the simulated brain running. Its neural activity is translated into motor commands that move the mechanical legs below. The catch: this is a computational model built from biological wiring. I engineered the interface connecting its outputs to the robot. First, the fly brain entered DOOM. Then it played Chrome Dino. Now it has a physical body. 166,700 neurons. 25 million connections. One nervous system walking through the real world.

Frank

4,732,965 views • 2 days ago

Fed district partisanship (and a few other vars) into the sugar and bitter taste neurons of a fruit fly. The fly predicts races by steering its body in a given direction and it manages to be 89% accurate on house races!

Fed district partisanship (and a few other vars) into the sugar and bitter taste neurons of a fruit fly. The fly predicts races by steering its body in a given direction and it manages to be 89% accurate on house races!

Ryan McComb

16,873 views • 3 days ago

now i trained the fly to play mario kart he's honestly really good (sound on)

now i trained the fly to play mario kart he's honestly really good (sound on)

Nick Walton

16,399 views • 4 days ago

so yeah I plug the fly brain data to simulate an autonomous driving model for Skypad, the viral Philippine transport vehicle real fly wiring + simulated neurons + assisted avoidance meet flypad autonomous driving cooked using openclaude harness with Astra

so yeah I plug the fly brain data to simulate an autonomous driving model for Skypad, the viral Philippine transport vehicle real fly wiring + simulated neurons + assisted avoidance meet flypad autonomous driving cooked using openclaude harness with Astra

Kevin

16,403 views • 4 days ago

Spread your wings in the wide sky Fly through your dreams Fly away, just fly away Fly higher Shine on me with a new light To the place where my dreams were engraved Fly away, just fly away Fly higher FLY WITH JINJIN AND MOONBIN #JINJIN_MOONBIN_Fly

Spread your wings in the wide sky Fly through your dreams Fly away, just fly away Fly higher Shine on me with a new light To the place where my dreams were engraved Fly away, just fly away Fly higher FLY WITH JINJIN AND MOONBIN #JINJIN_MOONBIN_Fly

Head empty, just Jinjin

24,650 views • 2 years ago

Scientists used 200,000 human brain cells to play DOOM. The cells had never seen a computer. They learned in real time. They got better. Separately: an entire fruit fly brain, 139,000 neurons, 50 million connections, was copied into a laptop. It predicts real brain responses with 90%+ accuracy. The biological computing timeline: 2021 → neurons learn Pong 2022 → fruit fly brain fully mapped 2024 → entire fly brain simulated on a laptop 2026 → human brain cells play DOOM A GPU needs ~700W to run an LLM. The brain runs the whole show on 20W. We're spending trillions on building the next giant AI training computer but what if the solution to our energy needs was just copying the technology already running in our heads?

Scientists used 200,000 human brain cells to play DOOM. The cells had never seen a computer. They learned in real time. They got better. Separately: an entire fruit fly brain, 139,000 neurons, 50 million connections, was copied into a laptop. It predicts real brain responses with 90%+ accuracy. The biological computing timeline: 2021 → neurons learn Pong 2022 → fruit fly brain fully mapped 2024 → entire fly brain simulated on a laptop 2026 → human brain cells play DOOM A GPU needs ~700W to run an LLM. The brain runs the whole show on 20W. We're spending trillions on building the next giant AI training computer but what if the solution to our energy needs was just copying the technology already running in our heads?

Josh Kale

136,752 views • 6 months ago

I’m excited to announce the world’s first Fly Language Model. It uses a new architecture called GPF (Generative Pre-trained Fly). It adapts a small pre-trained model with the actual fly connectome. Don't believe it's real? Try it out below, view the code, and read the paper. What will you ask the fly?

I’m excited to announce the world’s first Fly Language Model. It uses a new architecture called GPF (Generative Pre-trained Fly). It adapts a small pre-trained model with the actual fly connectome. Don't believe it's real? Try it out below, view the code, and read the paper. What will you ask the fly?

Alex Wormuth

92,242 views • 4 days ago

This might make you think twice next time you reach for the fly swatter. The whole brain wiring diagram of the fly! All 140,000 neurons and their 50M synapses are available to explore in Codex, the Connectome Data Explorer: #sfn23

This might make you think twice next time you reach for the fly swatter. The whole brain wiring diagram of the fly! All 140,000 neurons and their 50M synapses are available to explore in Codex, the Connectome Data Explorer: #sfn23

FlyWire

73,565 views • 2 years ago

(1/2) When Janelia set out to map every neuron in the fly brain in 2008, many thought it couldn’t be done. Nearly 20 years later, a complete wiring diagram of the 166,000+ neurons of the adult male fruit fly is helping to reveal the brain’s secrets:

(1/2) When Janelia set out to map every neuron in the fly brain in 2008, many thought it couldn’t be done. Nearly 20 years later, a complete wiring diagram of the 166,000+ neurons of the adult male fruit fly is helping to reveal the brain’s secrets:

HHMI

27,719 views • 12 days ago

For the first time in history, GPT-6 Astra created a fly that is simply a fly. Not a chatbot pretending to be a fly. Not a simplified animation. A digital fly running its own connectome brain, perceiving a real room through its own senses and interacting with the environment using the power of AR. The entire loop is surprisingly wild: Spectacles capture the retina, light rays, hands, and the surrounding room → the room is converted into an inventory → sensory information is fed into the fly’s neural system → the MaleCNS connectome processes it → descending neurons and motor pools generate actions → the commands return to the Spectacles → wings, legs, and head respond. In other words, the fly sees through its own eyes, processes the world through its own simulated brain, and acts through its own virtual body. We are moving from AI that generates outputs to AI that actually has a body, senses, a brain, and a closed perception-action loop. The 🪰 is no longer just simulated. It is experiencing a world The developer Pavlo Tkachenko is pinned below in the post

For the first time in history, GPT-6 Astra created a fly that is simply a fly. Not a chatbot pretending to be a fly. Not a simplified animation. A digital fly running its own connectome brain, perceiving a real room through its own senses and interacting with the environment using the power of AR. The entire loop is surprisingly wild: Spectacles capture the retina, light rays, hands, and the surrounding room → the room is converted into an inventory → sensory information is fed into the fly’s neural system → the MaleCNS connectome processes it → descending neurons and motor pools generate actions → the commands return to the Spectacles → wings, legs, and head respond. In other words, the fly sees through its own eyes, processes the world through its own simulated brain, and acts through its own virtual body. We are moving from AI that generates outputs to AI that actually has a body, senses, a brain, and a closed perception-action loop. The 🪰 is no longer just simulated. It is experiencing a world The developer Pavlo Tkachenko is pinned below in the post

Zentrix⌚️

313,652 views • 2 days ago

HOW TO BEAT DEADLOCK SNEAK PEEK OF A VIDEO IM WORKING ON! AIMING TO GET THIS ONE OUT QUICK, SHOWED THIS AT THE END OF STREAM!

HOW TO BEAT DEADLOCK SNEAK PEEK OF A VIDEO IM WORKING ON! AIMING TO GET THIS ONE OUT QUICK, SHOWED THIS AT THE END OF STREAM!

youreashotgun

17,198 views • 6 months ago

For the first time ever, a fruit fly brain has been trained to launch a memecoin. Not "AI inspired by a brain" — the real connectome. 165,122 neurons, 10,228,000 synaptic connections, every one traced from an actual male Drosophila by electron microscopy. Janelia FlyEM + Google Research, released CC-BY. How it works: Screenshots of the live launchpad go in through 892 retinotopic hex columns into L1 and L2 — the neurons sitting directly behind a fly's photoreceptors. All 165,122 integrate. The cursor comes back out of the descending neurons a fly actually walks with: DNa02's left-right asymmetry steers, DNa01 drives forward, DNp09 clicks. The wiring is never trained. Only synaptic gains — you don't get to invent connections the fly doesn't have. Proof its eye is doing real work: in light mode it filled 0 of 3 form fields. In dark mode, 2 of 3. Its retina is a luminance map, so invert the contrast and it goes blind. And the honest split: the fly types the description and the ticker. The rig does the name, the pair and the tax. CA: 0x4eb990547bce4a982432ca88cf5fae7eed1a2d35

For the first time ever, a fruit fly brain has been trained to launch a memecoin. Not "AI inspired by a brain" — the real connectome. 165,122 neurons, 10,228,000 synaptic connections, every one traced from an actual male Drosophila by electron microscopy. Janelia FlyEM + Google Research, released CC-BY. How it works: Screenshots of the live launchpad go in through 892 retinotopic hex columns into L1 and L2 — the neurons sitting directly behind a fly's photoreceptors. All 165,122 integrate. The cursor comes back out of the descending neurons a fly actually walks with: DNa02's left-right asymmetry steers, DNa01 drives forward, DNp09 clicks. The wiring is never trained. Only synaptic gains — you don't get to invent connections the fly doesn't have. Proof its eye is doing real work: in light mode it filled 0 of 3 form fields. In dark mode, 2 of 3. Its retina is a luminance map, so invert the contrast and it goes blind. And the honest split: the fly types the description and the ticker. The rig does the name, the pair and the tax. CA: 0x4eb990547bce4a982432ca88cf5fae7eed1a2d35

Fruit fly dev

756,578 views • 5 days ago