A male fruit fly's brain, mapped over nearly a decade by Google and the HHMI Janelia Research Campus, has been used in a simulation that claims to have taught it to play the card game Balatro with a 20 percent success rate. The connectome — a detailed map of the brain's neural connections — was published in the scientific journal Cell and contains over 166,000 neurons and 125 million synaptic connections. This makes it the largest brain map ever created and was originally intended for neuroscience research to better understand how neural networks function. According to Reddit user ActualAerie1011, the simulation involved coding an algorithm to identify the right "seeds," which are essentially starting conditions in a roguelike game. The algorithm and the fruit fly brain model then play the same seed. The system uses a reinforcement learning approach, rewarding the fly when it performs well and punishing it when it makes mistakes. This method is commonly used in artificial intelligence to train systems through trial and error. The reported success rate is 20 percent on the easiest difficulty level of the game. The user initially faced skepticism from online commenters, who questioned the validity of the claim and asked for more details. Under pressure, ActualAerie1011 explained that he had completed another personal project and was accused of cheating. He noted that he had not completed formal computer science studies and had learned to program in Python on his own. Additionally, he mentioned that he does not use GitHub to publish his work and works on his projects independently. The simulation uses the fruit fly connectome as a fixed reservoir, meaning the brain's structure is used as a static framework. The characteristics of the game Balatro are formatted in JSON, a common data format, and then processed through three mathematical functions called tanh updates. These updates simulate how the game features are absorbed and processed by the brain's neurons. The features are then grouped into 256 dimensions, which are used by a simple machine learning model to evaluate the actions of the "teachers" — the algorithm guiding the brain. The unique aspect of this simulation is the scale of the task, using a real animal connectome as a computational reservoir. Instead of using dopamine or punishment signals, the learning algorithm plays the game and finishes it, and the brain then mimics the algorithm's decisions. The efficiency of the algorithm is highlighted, as it successfully trains the brain to make similar choices.