AI models are terrible at betting on soccer—especially xAI Grok

Editor's take

AI models, including xAI's Grok, demonstrate a consistent inability to predict outcomes in English Premier League soccer matches, performing no better than random chance. This finding is significant as it highlights the current limitations of even advanced large language models in complex, real-world probabilistic forecasting, a domain where human intuition and specialized domain knowledge still hold a substantial advantage. It suggests that while LLMs excel at text generation and information synthesis, they lack the nuanced understanding of sports dynamics required for reliable betting predictions.

The failure of models like Google's Gemini, OpenAI's GPT-4, and Anthropic's Claude, alongside Grok, underscores a critical gap between their broad linguistic capabilities and specialized predictive prowess. Future developments to watch include whether fine-tuning on vast datasets of historical match data, incorporating live game statistics, or integrating external factors like player form and weather patterns can improve performance. It would be compelling to see if any model can consistently outperform a simple baseline, such as always betting on the home team, within a statistically significant margin.