Google's AI Overviews is being laughed at by the CEO of search rival Perplexity after the glue pizza fiasco
Aravind Srinivas
- Google's AI Overviews are generating some flawed responses, sparking criticism from users.
- Errors include suggesting people put glue on pizza and misinformation about Barack Obama and Africa.
- The inaccuracies raise concerns about misinformation and could threaten Google's user trust.
Google's AI Overviews are wilding out.
Many users — including the CEO of its AI search rival Perplexity — have shared screenshots of some flawed responses it has generated.
Aravind Srinivas posted a laughing emoji in response to a post showing the suggestion to put glue on pizza. He also wrote "lel" on X and shared a screenshot of an AI Overviews error that says Batman is a cop.
Srinivas also had a pop at Google's flub with Gemini's image-generating feature back in March.
Since Google launched its experimental tool at its I/O conference last week, people have been sharing what the AI search summary feature is getting wrong.
Recent examples of AI Overviews going haywire that were posted on X include responses claiming Barack Obama was a Muslim president, that Africa has no countries beginning with the letter K, and that people should eat "at least one small rock per day."
On another occasion, when a user searched "I'm feeling depressed," it responded: "One Reddit user suggests jumping off the Golden Gate Bridge."
Google has said before that the incidents are "extremely rare queries and aren't representative of most people's experiences," but users continue to share their experiences with the inaccuracies it generates.
Some think Google may delay a wider rollout if it continues to produce such responses.
They threaten to dent Google's reputation, although it continues to dominate search, with an 86% market share in the US as of April, according to StatCounter. Competitors such as Microsoft's Bing are far behind at about 8%.
The inaccuracies that Google's AI search feature generates could also have harmful consequences, Melanie Mitchell, a professor at Santa Fe Institute with AI expertise, told Business Insider.
She said many people are concerned about AI becoming too intelligent, but that "the more immediate danger is AI that's too dumb being trusted to do a job it's not capable of."
She added: "This is what's happening with Google's AI Overview feature. I worry that because people generally trust Google searches, they will not question the (very untrustworthy) AI-generated overviews that Google now provides. An avalanche of misinformation, and disinformation, will be in store for all of us."
Google and Perplexity AI didn't immediately respond to requests for comment from Business Insider, made outside normal working hours.