Google, Microsoft's AI Models Misfires on Super Bowl Predictions, Revealing Flaws in GenAI Logic

As further proof that GenAI is prone to fabrication, Google's Gemini chatbot, formerly known as Bard, believes the 2024 Super Bowl has already transpired. It even has the (fictional) statistics to back it up.

Google's Gemini Chatbot Fabricates 2024 Super Bowl Results, Showcases AI Quirks

TechCrunch said that according to a Reddit thread, Gemini, powered by Google's GenAI models of the same name, is responding to queries regarding Super Bowl LVIII as if the game concluded yesterday or weeks ago. Like many other bookmakers, it appears to favor the Chiefs over the 49ers (sorry, San Francisco supporters).

Gemini embellishes quite imaginatively, providing a player statistics breakdown in which Patrick Mahomes, quarterback of the Kansas Chiefs, ran 286 yards for two touchdowns and an interception, as opposed to Brock Purdy's 253 yards and one touchdown.

It is not solely Gemini. Similarly, the Microsoft Copilot automation asserts that the game has concluded and supports this assertion with inaccurate citations. The 49ers, not the Chiefs, were declared the winners "with a final score of 24-21," which may indicate a San Francisco bias.

Copilot operates on a GenAI model that is virtually indistinguishable from the model that forms the foundation of OpenAI's ChatGPT (GPT-4). However, during the testing, ChatGPT avoided repeating the error.

AI Chatbots Tackle Fictional Super Bowl, Highlighting GenAI's Reality Check Need

It is all rather frivolous, and the issue may have been resolved by now, considering the reporter's inability to reproduce the Gemini responses in the Reddit thread. However, it also exemplifies the significant shortcomings of contemporary GenAI and the perils associated with overconfidence.

There is no malice on the part of the LLMs. They need better intent and find the concepts of truth and falsehood devoid of significance. They have merely acquired the ability to link particular words or expressions with specific ideas, notwithstanding the fallibility of those associations. Therefore, Gemini and Copilot's claims regarding the Super Bowl in 2024 and 2023 are unfounded.

Like most GenAI vendors, Google and Microsoft openly recognize that their GenAI applications are not flawless and are, in fact, susceptible to errors. However, these acknowledgments, in the form of small print, are easily overlooked.

Super Bowl disinformation is by no means the most detrimental instance of GenAI deviating from its intended path. Endorsing torture, perpetuating ethnic and racial stereotypes, or writing convincingly about conspiracy theories are likely the defining characteristics of this individual. Nonetheless, it serves as a practical reminder to verify assertions made by GenAI programs. They may not be accurate.

Photo: Pawel Czerwinski/Unsplash