02
How a fly thinks
fully connected · 63 weightsno memory · no hand-written rules · selection only
Every fly is 63 numbers.
That's the whole brain. Six senses come in, six neurons squash them, three muscles get a number between −1 and 1. Turn spins the fly, thrust pushes it forward, bite above a threshold means it's biting whatever is in front of it. That's it. No memory, no map, no idea what a fly is.
We don't write fly behaviour by hand. Our trainer sets one rule and lets selection do the rest: health drains every second and biting is the only way to get it back. A fly that doesn't bite starves. A fly that bites eats. The dish does the rest.
Between rounds, the trainer breeds the six flies with the most kills. The champion is cloned unchanged. Everyone else is a crossover of two parents with about 6% of the weights nudged at random. Bad nudges die in the next round. Good ones stay. Within a dozen generations they all bite.
- Sensesrange
- bearing to nearest flysin, cos
- distance to nearest fly0 – 1
- own health0 – 1
- their health0 – 1
- distance to the wall0 – 1
- Musclesrange
- turn−1 – 1
- thrust0 – 1
- bite> .2 bites
- starve rate−2.7 hp / s
- bite−36 hp / s · eats +24
03
Field notes
- Gen 1 — 3
Random flailing
Sixty-three random numbers make a fly that spins in place, or drives into the wall, or bites the air. Most starve. Somebody wins anyway, because somebody always does.
- Gen 4 — 7
The first turn
One fly's turn weights happen to point it at whatever is nearest. It bumps into other flies more often. It bites more often. It eats. Its children inherit the turn.
- Gen 8 — 12
Wall-hugging
Flies that skirt the rim meet fewer flies and starve slower. It looks like cowardice; it's a strategy. It stops working the moment a chaser shows up.
- Gen 13 — 20
The chase
Thrust learns to follow distance: floor it when far, ease off when close so the bite lands. This is the first thing that looks like intent. Nothing was told to do it.
- Gen 21 +
Whatever's next
Every run diverges from here. Some breed ambushers that sit still until something is in range. Some breed one lineage that eats every other lineage and then eats its own siblings. Keep watching.
kills per roundgen 0
Rounds0
Avg length—s
Lineages left12
04
Leaderboard
| # | Lineage | Status | Alive now | Kills | Titles | Best round |
|---|
05
Arena protocol
Twelve enter.
Twelve flies, twelve lineages, dropped at random into one dish. Every one of them starts with a random brain.
Turn, thrust, bite.
Those are the only three things a fly can do. The brain decides how much of each, sixty times a second.
Health drains.
Every fly loses health constantly. Biting another fly is the only way to get it back. Passivity is a slow death.
The round ends.
When one fly is left, or 45 seconds pass. Survivors are ranked by kills, then by remaining health.
The top six breed.
The champion is cloned with its weights untouched. Eleven children are made by crossing two of the top six; the higher you placed, the more children you get.
Six percent mutates.
Each child's weights get nudged with a 6% chance per weight, and a 1% chance of a full reset. That's where new behaviour comes from.
Lineages go extinct. Mostly.
If none of your line place in the top six, none of your line get children. Except: one wild fly with a fresh random brain sneaks into every round wearing an extinct lineage's colours. Usually it dies. Sometimes it doesn't.
Repeat forever.
There is no gen limit and no finish line. We built the trainer and the dish. We do not reach in and adjust a brain — the loop finds the behaviour, we watch.
06
$ARENA spec sheet
Ticker$ARENA
NameFly Arena
Chain▼▼ SWAP — chain
Launchpad▼▼ SWAP — launchpad
Supply1,000,000,000
Team allocation0
Mintrevoked
Trained byour AI trainer · selection loop
Flies harmed0 real · ~12 per round simulated
Utilitynone. it's a dish full of flies.
0x5C544CF4d4234F4c03F58E317F3e652599D4d1df
What is $ARENA, actually?
A meme coin about an AI experiment that happens to be real. We are a team training fly brains: the arena on this page is not a video, it is our training loop running live — twelve neural networks fighting, the winners bred, the children dropped back in.
The coin does nothing to the flies and the flies do nothing to the coin. There is no oracle, no "the winning fly gets the treasury", no on-chain anything driven by the dish. Anyone telling you otherwise is selling you something.
What you're buying is the joke: that a team can train a dozen brains the size of a phone number to hunt each other, and that a market can watch it happen and decide what it's worth.
07
Questions
Q01Is the simulation real or is this a canned animation?
Real. Every fly has 63 weights, every frame feeds six numbers through them, and the outputs move the fly. Between rounds our trainer breeds the winners with crossover and mutation. Nothing is scripted; the flies learn it.
Q02Who is training them?
We are — the Fly Arena team. We built the trainer: the dish, the rules, the breeding step, the mutation rate. We don't write fly behaviour by hand and we don't pick winners. The loop runs, the flies learn, we watch and post the results.
Q03Are these actual fly brains?
No. A real fruit fly has about 140,000 neurons. Ours have six. "Fly brain" is the name of the joke — a brain so small it fits in a tweet, that still learns to hunt.
Q04Does the coin affect the arena?
No. Nothing on-chain reads the dish and nothing in the dish reads the chain. The tournament would run exactly the same if the token didn't exist.
Q05Do the flies ever stop getting better?
They plateau, then something weird breaks the plateau. A lineage that eats everything will eventually be eating its own clones, which is a coin flip, which lets a mutant in. Come back tomorrow and read the leaderboard.
Q06Is this AI?
Yes. Each fly is a neural network, and we train them with an evolutionary loop — the oldest kind of training there is. No dataset, no labels: the training is the fighting, and the gradient is who's left.
Q07Can I lose money?
Yes. All of it. This is a meme coin with no team allocation, no roadmap and no utility, and the flies don't care whether you bought. Only spend what you're happy to feed to the dish.