Autonomous AI Weather Trading and Prediction Battle Results

AI Prediction Battle Results

The author evaluated the performance of two LLMs, GPT-4 and Claude 3 Opus, in a prediction battle focused on weather events. In the first week, the models were tested on two specific events. Claude 3 Opus outperformed GPT-4 by correctly predicting a lowest temperature value, while GPT-4 missed the target. The author intends to expand the scope of these battles in future weeks to include four to five events per session to better measure model accuracy.

Autonomous Trading and Machine Learning Experiments

The author is developing an autonomous system called QuantWFX for weather-based trading on Polymarket. The system utilizes dynamic expected value calculations to manage positions autonomously. While initial testing phases resulted in losses due to parameter misconfiguration, the system has begun to show a positive recovery trend as it runs autonomously.

Additionally, the author is building an autonomous research platform designed to identify trading strategies by analyzing data from platforms like Polymarket and Hyperliquid. This platform has completed 125 experiments, currently outperforming the market baseline by 0.26%. The author plans to further refine these machine learning models by testing them against unseen data and leveraging LLMs to accelerate data cleaning and training code generation.