When giant artificial intelligence models attempted to bet on Premier League matches, the outcome was a disaster. All the giants, from Google to OpenAI, exhausted their virtual funds in the KellyBench test and declared bankruptcy.
The much-praised artificial intelligences, capable of writing complex code in seconds and acing exams, hit a wall when they stepped onto the green field. The KellyBench experiment, conducted by the startup General Reasoning, proved that while AI is successful with static data, it is equally helpless in the chaotic flow of the real world. Eight giant AI models, having memorized all the statistics of the English Premier League's 2023-2024 season, gradually squandered their fortunes at the virtual betting table.
The House Always Wins
The experiment results were a complete disappointment for the tech giants. Elon Musk's xAI Grok 4.20 model threw in the towel and declared bankruptcy early on. Google's Gemini 3.1 Pro model, on the other hand, exhibited a true gambler's profile; while it surprised everyone by making a 34% profit in its first attempt, it went bust by zeroing out its entire bankroll in the second attempt. The model that escaped with the least loss was Anthropic's Claude Opus 4.6; however, even it couldn't avoid an average loss of 11%.
Researchers state that this situation stems from AI's inability to analyze the complexity of real life and unexpected events. In other words, it appears that no matter how skilled AI is at writing code, it is not yet as capable as a human in predicting the chaos a last-minute goal in the Premier League would create.
0 Comments: