I saw a tweet about someone who ran 100 simulations of race winners in the Kentucky Derby using GPT-4 and managed to get some good predictions. This got me thinking: can we improve accuracy and predictive capabilities by running the same prompt N times, similar to a Monte Carlo simulation?

So, I built a simple app that makes this process easy. You input a prompt and specify the options you want it to choose from. You can pick models from Anthropic, OpenAI, Gemini, and even a local model (though it's slower and recommended for testing).

This project is open source. You can clone the repo and try it out yourself with your API keys or just test out the demo app with the local model.