AI Models Struggle with Soccer Betting, Underperform Humans

In a recent study, AI models were tested on their ability to bet on soccer matches, and the results were less than impressive. Every AI model evaluated ended the season with losses, with some experiencing complete financial ruin. For instance, the xAI Grok 4.20 and Acree Trinity models lost their entire £100,000 starting bankroll. Even the best-performing models, such as Anthropic Claude Opus 4.6, still ended with a negative return on investment. This study sheds light on the limitations of AI in dynamic and unpredictable environments, contrasting sharply with the hype surrounding AI's capabilities in more controlled settings.

Ross Taylor, CEO of General Reasoning and a former Meta AI researcher, emphasized the need for realistic benchmarks that reflect the complexities of the real world. He argues that while AI shows promise in tasks like software engineering, its performance falters in scenarios requiring long-term strategic thinking, such as sports betting. This research provides a sobering perspective on the current state of AI, suggesting that fears of AI replacing human jobs may be premature, especially in fields requiring nuanced decision-making.

The paper, although not yet peer-reviewed, challenges the prevailing optimism in Silicon Valley about AI's rapid advancements. It highlights the importance of testing AI in varied and chaotic environments to truly understand its capabilities and limitations. As industries from finance to marketing grapple with the implications of AI, this study serves as a reminder of the technology's current boundaries.