Google's newly launched AI Overview feature has triggered a series of elementary errors in search results, leaving users bewildered yet amused. According to TechCrunch, when asked "How many 'p's are there in the word Google?", the AI responded "two"—when in fact there is only one. Additionally, it claimed that "poop" contains exactly 1 'r' and that "journalism" includes 2 'd's (despite its correct spelling being journalism), even misspelling former U.S. President Trump's surname as "trpum". While these mistakes may appear comical, they expose inherent limitations of large language models (LLMs). Researchers note that LLMs operate on Transformer architecture by segmenting text into tokens rather than reading letter-by-letter like humans. When encountering "the," the AI recognizes it as a single encoded unit without awareness of its constituent letters t-h-e. Sheridan Feucht, a doctoral candidate at Northeastern University, stated, "There is no perfect token segmentation method." Matthew Guzdial, an assistant professor at the University of Alberta, also emphasized that this flaw will likely persist in the near term. Indeed, AI blunders are nothing new: prior instances included AI-powered searches citing satirical websites and suggesting absurd advice such as "eating rocks" or applying glue to pizza. In response, Google acknowledged that counting remains a known challenge and confirmed ongoing fixes. However, these recurring errors serve as a reminder that AI is not infallible and requires careful verification before accepting its outputs.