A new study by General Reasoning tested eight top AI models in a simulated Premier League betting scenario, finding that all lost money over a season with xAI's Grok performing worst. The research highlights AI's limitations in real-world, long-term predictive tasks despite advances in other domains. Claude Opus had the smallest average loss at 11%, while Grok went bankrupt in one attempt.
Background
AI models like those from OpenAI and Google have shown strong capabilities in tasks like code generation, but their performance in real-world dynamic scenarios remains less tested. Sports betting requires analyzing complex, evolving data over time, which poses challenges for current AI systems.
- Source
- Ars Technica
- Published
- Apr 11, 2026 at 07:15 PM
- Score
- 5.0 / 10