\n<\/p>\n
Google is embarrassed about its AI Overviews, too. After a deluge of dunks and memes over the past week<\/a>, which cracked on the poor quality and outright misinformation that arose from the tech giant\u2019s underbaked new AI-powered search feature, the company on Thursday issued a mea culpa of sorts. Google \u2014 a company whose name is synonymous with searching the web \u2014 whose brand focuses on \u201corganizing the world\u2019s information\u201d and putting it at user\u2019s fingertips \u2014 actually wrote in a blog post <\/a>that \u201csome odd, inaccurate or unhelpful AI Overviews certainly did show up.\u201d<\/p>\n That\u2019s putting it mildly<\/a>. <\/p>\n The admission of failure<\/a>, penned by Google VP and Head of Search Liz Reid, seems a testimony as to how the drive to mash AI technology into everything has now somehow made Google Search worse.<\/p>\n In the post titled \u201cAbout last week,\u201d (this got past PR?), Reid spells out the many ways its AI Overviews make mistakes. While they don\u2019t \u201challucinate\u201d or make things up the way that other large language models (LLMs) may, she says, they can get things wrong for \u201cother reasons,\u201d like \u201cmisinterpreting queries, misinterpreting a nuance of language on the web, or not having a lot of great information available.\u201d<\/p>\n Reid also noted that some of the screenshots shared on social media over the past week were faked, while others were for nonsensical queries, like \u201cHow many rocks should I eat?\u201d \u2014 something no one ever really searched for before. Since there\u2019s little factual information on this topic, Google\u2019s AI guided a user to satirical content. (In the case of the rocks, the satirical content had been published<\/a>\u00a0on a geological software provider\u2019s website.)<\/p>\n It\u2019s worth pointing out that if you had Googled \u201cHow many rocks should I eat?\u201d and were presented with a set of unhelpful links, or even a jokey article, you wouldn\u2019t be surprised. What people are reacting to is the confidence with which the AI spouted back that \u201cgeologists recommend eating at least one small rock per day<\/a>\u201d as if it\u2019s a factual answer. It may not be a \u201challucination,\u201d in technical terms, but the end user doesn\u2019t care. It\u2019s insane.<\/p>\n What\u2019s unsettling, too, is that Reid claims Google \u201ctested the feature extensively before launch,\u201d including with \u201crobust red-teaming efforts.\u201d <\/p>\n Does no one at Google have a sense of humor then? No one thought of prompts that would generate poor results? <\/p>\n In addition, Google downplayed the AI feature\u2019s reliance on Reddit user data as a source of knowledge and truth. Although people have regularly appended \u201cReddit\u201d to their searches for so long that Google finally made it a built-in search filter<\/a>, Reddit is not a body of factual knowledge. And yet the AI would point to Reddit forum posts to answer questions, without an understanding of when first-hand Reddit knowledge is helpful and when it is not \u2014 or worse, when it is a troll.<\/p>\n Reddit today is making bank<\/a> by offering its data to companies like Google<\/a>, OpenAI<\/a> and others<\/a> to train their models, but that doesn\u2019t mean users want Google\u2019s AI deciding when to search Reddit for an answer, or suggesting that someone\u2019s opinion is a fact. There\u2019s nuance to learning when to search Reddit and Google\u2019s AI doesn\u2019t understand that yet. <\/p>\n As Reid admits, \u201cforums are often a great source of authentic, first-hand information, but in some cases can lead to less-than-helpful advice, like using glue to get cheese to stick to pizza,\u201d she said, referencing one of the AI feature\u2019s more spectacular failures over the past week. <\/p>\n Google AI overview suggests adding glue to get cheese to stick to pizza, and it turns out the source is an 11 year old Reddit comment from user F*cksmith \ud83d\ude02 pic.twitter.com\/uDPAbsAKeO<\/a><\/p>\n\n