Après Paul le Poulpe, Claude, ChatGPT et Gemini s'affrontent : Découvrez l'équipe sacrée championne du monde selon les IA
Claude and OpenAI predict a victory for Spain, Mistral favors the French, while DeepSeek and Qwen bet on Argentina: predictions made by artificial intelligence on the outcome of the World Cup are multiplying, at the very moment when a team of German researchers is preparing to scientifically evaluate the predictive capabilities of these different models.
The 2026 World Cup, which began this Thursday with a victory for Mexico against South Africa (2-0), is seeing the use of artificial intelligence tools become widespread among fans and financial institutions in an attempt to anticipate the results of the competition.
Bank of America analysts found that Copilot, the model developed by Microsoft, placed Spain on par with France. Similar results were obtained by the technology news site Tom's Guide, which surveyed Google's Gemini, OpenAI's ChatGPT, and Perplexity. Their queries generally showed Spain leading in the algorithmic rankings, closely followed by France. Meanwhile, the news site Decrypt observed that the Chinese competitors DeepSeek and Qwen both predicted Argentina would win.
This trend towards non-human predictions is reminiscent of the phenomenon of Paul the Octopus who, during the 2010 tournament, designated the winning teams by choosing his food from containers marked with national flags.
The widespread use of these technologies is a major first for a World Cup finals tournament. While ChatGPT's public launch in November 2022 coincided with the very beginning of the World Cup in Qatar, the chatbot had not yet gained mainstream popularity or reached its current level of adoption.
Faced with this enthusiasm, researchers at Ludwig-Maximilian University (LMU) in Munich, Germany, are trying to determine which model will prove most accurate in its analyses. They are evaluating the accuracy of each model in real time on a public platform called LLM SoccerArena.
Researcher Stefan Feuerriegel explained in a press release the relevance of this approach, stating that the scientific community needs benchmark tests that not only evaluate abstract tasks, but also how models handle constantly evolving information, uncertainty, and concrete results that can then be directly compared to real-world situations. The university team is also testing the AI's ability to integrate evolving data gathered online into its predictions in real time, such as players' health and injuries, official team lineups, and fluctuations in the sports betting market.
AFP
Commentaires (1)
Participer à la Discussion
Règles de la communauté
💡 Astuce : Utilisez des emojis depuis votre téléphone ou le module emoji ci-dessous. Cliquez sur GIF pour ajouter un GIF animé. Collez un lien X/Twitter, TikTok ou Instagram pour l'afficher automatiquement.
Se connecter
Commentez avec votre profil, votre photo, et soyez averti des réponses.