In the world of artificial intelligence, Google's Gemini and Microsoft's Copilot have raised a few eyebrows by fabricating statistics about the Super Bowl of 2024, which has yet to take place. According to a thread on Reddit, Gemini, an AI model created by Google, answers queries about Super Bowl LVIII as if the match had already concluded.
Being somewhat imaginative, Gemini divulged detailed but made-up player statistics, for instance, claiming Patrick Mahomes, the Kansas City chief quarterback, pulled off two touchdowns and an interception for 286 running yards. Additionally, it predicted a Chiefs' win over the 49ers, likely to dishearten their San Francisco supporters.
Akin to its Google counterpart, Microsoft's Copilot insists on a game conclusion as well, providing false evidence for its claim. However, seeming to favor San Francisco, it posits the 49ers as the winners with a scoreline of 24-21. Copilot operates using a GenAI model very much like the one used by OpenAI’s ChatGPT.
Interestingly, despite being powered by a similar GenAI model, ChatGPT emerged as the better of the three by not committing the same blunder. At the same time, these bizarre predictions might serve as indicators of the glaring limitations of today's AI models, highlighting why it can be risky to place complete trust in them.
While GenAI’s 'make-believe' Super Bowl predictions are far from harmful, they offer a timely reminder to verify facts presented by GenAI bots. As these models rely on probabilities based on data fed to them, they may generate anything from nonsensical to downright false content, propagating inaccuracies from their training data.