The experiment is based on the scientific brain map of the fruit fly Drosophila MaleCNS v1.0. The brain has 166,700 neurons and 25.6 million synaptic connections, explained the programmer. Public Coinbase prices for the $BTC-$USDC pairs are converted into an RGB graph, which is fed to 3,335 brightness inputs and 811 color inputs of the R8 neural network. Purchases are encoded as 'sweet,' and sales as 'bitter.' A fixed neural reader converts brain activity into a command: buy, sell, or hold. Then, a middleware service between the neural network outputs and the real trading system checks limits and places spot orders through Coinbase Advanced.
The project operates in safe mode, without real money. The prices for the $BTC-$USDC pair were real, but the fly traded with virtual $100. Trading with real money is also possible: for this, according to the programmer, one can create a separate portfolio with a balance of no more than 100 $USDC and generate an API key with viewing and trading rights, but without withdrawal permissions. The neuro-fly, the developer advises, is capable of executing trades worth a maximum of $10 including fees, with no more than 24 such trades per day. The model cannot use leverage, open short positions, and if the portfolio drops by $20, new orders are blocked.
Moreover, the fly learns trading gradually: with a positive portfolio result, 15 dopaminergic PAM11 neurons are stimulated; with a loss, two aversive PPL101 neurons are stimulated, causing the fly to develop an avoidance reaction to such trades. However, demonstrating that the fly's brain learns to trade profitably has not yet been possible, admitted Wormuth. Synaptic changes by themselves do not prove the ability for profitable trading—this is a research experiment, not a ready-made strategy.
Earlier, a bot on the Solana blockchain managed to turn 0.227 $USDC into approximately $696,000 in a few seconds using a single transaction. The profit was made thanks to a sharp drop in the price of the meme coin ANB.
end-content
