2 hours ago
3 ARTICLE AD
Google simply cannot escape the drama related to its Super Bowl Sunday ad campaign that is supposed to promote its Gemini AI model. What was originally thought to be a bit of hallucinated information from Gemini—which would be bad all on its own—turns out to have not even come from Gemini at all. Sure, AI chatbots tend to be plagiarism machines, but this looks like it’s entirely on Google.
Just to catch you up: Google’s planned ad buy for this year’s Big Game includes 50 separate stories highlighting small businesses in all 50 states that have used Gemini tools to help their operations. The Wisconsin ad centers on a cheese store called the Wisconsin Cheese Mart. In the ad, it’s suggested that the company used Gemini to generate copywriting for its website. That copy included the claim that gouda makes up “50 to 60 percent of the world’s cheese consumption,” which is not true. Google took heat for this and eventually changed the text that appears in the advertisement to exclude that incorrect factoid about the cheese.
Now, as it turns out, that incorrect copy wasn’t even generated by Gemini despite the advertisement suggesting that it was. Thanks to the Internet Archive, we can see that the text originally purported to be generated by Gemini has been on the Wisconsin Cheese Mart website as far back as 2020.
So, good news: Gemini isn’t responsible for the factual error in the website copy. Bad news: Gemini isn’t responsible for any of the website copy despite Google seemingly claiming that the business used the AI tool to generate the text.
Making this all the more embarrassing is the fact that a Google executive publicly went to bat for the original, not-AI-generated text. Jerry Dischler, the President of Cloud Applications at Google Cloud, insisted in a back-and-forth on Twitter that the text was “not a hallucination” and that “Gemini is grounded in the Web.” That may be true, but the evidence suggests that this particular example wasn’t grounded in anything because it wasn’t from Gemini in the first place.
So now Google is in quite an awkward position. It defended its AI model for sharing false information, only for it to be revealed that the AI model didn’t even generate the text. And the company was ready to spend millions of dollars to advertise the functionality of its AI suite with examples that aren’t even actually from the tool itself. We’re like one step away from this ad going the route of old video game trailers that would show some extremely polished-looking promotional clip with the disclaimer “not actual gameplay footage” below it. “Sure, this isn’t AI-generated, but imagine if it was!”
