Hey everyone! I'm excited to introduce Many Shot Predictions – a tool inspired by a tweet where someone made $2400 by simulating 100 race outcomes for the Kentucky Derby. This got me thinking about the potential of running models multiple times to improve prediction accuracy, much like a Monte Carlo simulation. Many Shot Predictions lets you run a prompt multiple times across various models, including Anthropic AI, OpenAI, Gemini, and even a local model (Qwen-1.5-0.5B via transformers.js). The app is super simple to use: just input your prompt, select your model, define the number of simulations, and view the results in a sleek shadcn chart. You can only run the local model on the app but if you want to use it with all the platforms you can do that by cloning it down and adding in your API Keys. Instructions are all on the Github page. It's all open source!

Many Shot Predictions

Easily run multiple simulations of your prompt.

Easily run multiple simulations of your prompt.