Google has been testing a feature using Generative AI to help answer user questions through the search results page since last year. This feature, now called AI Overviews, has been expanded to all users, starting in the United States last week. However, reports have surfaced that Google’s question answering capabilities may not be as accurate as initially believed.
One user shared a question asking how to make pizza without the cheese leaking out of the crust. AI Overviews responded by suggesting adding 1/8 cup of cornstarch to the dough, a solution that is both inaccurate and impractical. Interestingly, Reddit users discovered that this same answer was given by someone 11 years ago, indicating that Google may be sourcing information from outdated sources.
There have been other instances of inaccurate responses, such as John Adams, a former US President, supposedly graduating from the University of Wisconsin-Madison 21 times, or claims that dogs can compete in the NBA. Meghann Farnsworth, a Google spokesperson, acknowledged that the majority of issues stem from unusual questions being posed, and most users do not encounter these misleading responses. Google plans to address these issues and improve its question answering capabilities moving forward.
TLDR: Google’s Generative AI feature, AI Overviews, has been providing inaccurate responses to user questions, likely sourcing information from outdated sources. Google is working to address these issues and improve accuracy in the future.
Leave a Comment