Google AI Overviews were announce a twain of week ago atGoogle I / O , and they ’ve already proven to be rather controversial . The intention to providehigh - quality answersto your questions summarized from the web , but a series of recentX ( formerly Twitter ) threadsshow how big of a fail it ’s already proven to be .
The response that went viral involve a very dubious pizza pie formula . As report , when actuate Google for an answer to the military issue of “ cheese not sticking to pizza pie , ” the AI Overview suggests adding nontoxic glue to your pizza to preclude the cheese from slide off . The exact Logos the AI overview give are as postdate : “ you could also impart about 1/8 cup of non - toxic gum to the sauce to give it more tackiness . ” Where did theGoogle AI overviewget the information as a source ? An 11 - year - sometime Reddit commentfrom this thread , in what was intelligibly a gag .
https://t.co/W09ssjvOkJpic.twitter.com/6ALCbz6EjK
& mdash ; SG - r01 ( @heavenrend)May 22 , 2024
The words “ cheese not stick to pizza ” generated this unexpected and suspicious response , and the internet is throw a study day with it . The Google AI Overview response has since gone viral , withsomeone even trying glue pizzajust to make the point .
It should be mark that we ’ve see a monolithic uptick in Reddit and meeting place posts establish up higher in Google searches . It ’s also deserving noting that Reddit recentlysigned a $ 60 million deal to let Googletrain its models on Reddit content . It ’s not hard to connect the Lucy in the sky with diamonds on how this might have happened .
It ’s not just Reddit though . AnotherAI Overview was posted onlinewith an answer to “ how many rock should I exhaust each day , ” which pulls information directly from The Onion .
herpic.twitter.com/FGbvO923gk
& mdash ; follow @bencollins on bluesky ( @oneunderscore__)May 23 , 2024
Part of the problem is the absolute conviction in which AI Overviews delivers its answers . It does n’t bestow up a link to an Allium cepa article and permit you do the adjudication . Instead , it treats every source like it ’s Wikipedia and delivers information in pure trust .
Google claims that its AI Overviews give user high - tone selective information and that such misplay are uncommon . Here is the official reply put up to Digital Trends by Google : “ The Brobdingnagian majority of AI Overviews provide high quality selective information , with golf links to dig deeper on the web . Many of the examples we ’ve seen have been uncommon queries , and we ’ve also seen examples that were doctored or that we could n’t reproduce . We conducted extensive examination before launching this novel experience , and as with other features we ’ve launched in Search , we treasure the feedback . We ’re taking fleet natural process where appropriate under our content policies , and using these examples to develop broader improvements to our systems , some of which have already started to roll out . ”
It seems as if some of these AI Overviews will be hard to correct if they ca n’t be regurgitate . The “ broader betterment ” will need to be the solution , and since Google allege they ’re already in the works , hopefully we ’ll begin to see some better searches presently . We ’ll have to see if Google responds to the site further beyond that financial statement . After welcome negative feedback around its simulacrum generation in Gemini earlier this year , Google had to excuse and pull it downto fix its issue .
For now , though , these are both honorable reminders of how measured we postulate to be when trusting AI engines for information . Google AI Overview startedrolling out to everyone in the U.S. in the beginning this calendar month , and with more nation issue forth soon . But with answers like this , there may be more peoplereaching for a style to turn over it offthan Google expect .
