Fruit Fly AI Beats Game 20% Of Time

A Redditor named ActualAerie1011 claims to have trained Google's fruit fly brain simulation to play the popular game Balatro, with a reported 20% win rate on the easiest difficulty setting. The simulation was trained using a custom seed-finding algorithm and rewards-based training, where the model was rewarded when it performed well and penalized when it didn't.

The model played the game with the default Red Deck on its lowest difficulty setting and managed to beat it, according to a video posted by ActualAerie1011 on Reddit. The custom seed-finding algorithm was used to identify seeds likely to produce big scores, as Balatro's runs are randomized.

Google recently released a complete map of an adult fruit fly's brain, which has been used to create the simulation. The Redditor claims that they are still training the model, so its performance could improve over time.

However, some users in the Reddit thread have expressed skepticism about the claim, mainly because ActualAerie1011 hasn't posted any technical proof or a GitHub repository for others to review and verify.

## Why it matters The claim, if true, demonstrates the potential of using brain simulations to play complex games like Balatro. The game, which combines elements of poker and solitaire, requires strategy and decision-making skills. If a fruit fly brain simulation can be trained to play the game, it could have implications for the development of artificial intelligence and machine learning.

The use of a custom seed-finding algorithm and rewards-based training also highlights the potential of these techniques in training AI models. As the field of AI continues to evolve, we can expect to see more innovative applications of brain simulations and machine learning algorithms.