Can an LLM price a World Cup match better than the market (and make money)? (christian.bock.bio)

A recent experiment tested whether a large language model (LLM) could outperform market odds in predicting soccer match outcomes, specifically during the 2026 World Cup. The researcher utilized an automated system that processed match data to generate probabilities for different outcomes, which were then compared against betting market prices. Over 121 bets, this method yielded a substantial profit, increasing the initial bankroll from $255 to $771, despite winning only 51.2% of the bets. This outcome underscores the potential for sophisticated AI models to develop betting strategies that exploit market inefficiencies. The significance of this development lies in the model's ability to reason through complex inputs—such as team statistics, player form, and match context—without relying on initial market odds. By leveraging a modified Poisson distribution to estimate potential scorelines, the model was able to provide calibrated probabilities for a wide range of betting markets. This approach not only highlights the value of integrating AI with betting strategies but also suggests a broader application of LLMs in financial forecasting and decision-making in uncertain environments. The reliance on a robust methodology for evaluating risk and potential returns, such as the fractional Kelly criterion, further adds a layer of sophistication and reliability to AI-driven financial models.