Google is doubling down on making generative artificial intelligence the centerpiece of its classic search engine. However, the technology still trips over the absolute basics. Despite dramatic accuracy improvements over the last two years, users on social media are having a great time pointing out that Google’s AI Overviews still cannot handle simple spelling tests.
A hilarious string of spelling errors on Google Search’s AI Overviews
The latest round of internet mockery started when users began testing the search tool with straightforward vocabulary questions. When asked how many “e”s are in the word “astronomical,” Google’s AI Overview confidently stated there are exactly two, spelling it out as “a-s-t-r-e-n-o-m-i-c-a-e-l”.
Multiple reports make it clear that this is far from an isolated glitch. The system has claimed there are two “p”s in the word Google, exactly one “r” in poop, and two “d”s in journalism—which it creatively spelled as “j-o-u-r-n-a-d-i-s-m.” It even managed to identify the single “p” in a former U.S. president’s last name, only to spell the actual name as “t-r-p-u-m”. Google acknowledged the issue in a statement to TechCrunch, noting that counting characters within words is a known challenge for large language models (LLMs) and that teams are working on a fix.
Why smart brains are bad at letters
It seems bizarre that an advanced AI platform capable of writing complex code or solving long-standing mathematical riddles flunks basic spelling. However, the issue stems directly from how these models process language.
As researchers explained, AI doesn’t see sentences as strings of letters like humans do. Instead, modern software uses a so-called “transformer architecture.” This approach breaks text into chunks called “tokens.” A token can be a whole word, a single syllable, or a part of text.
Matthew Guzdial, an AI researcher at the University of Alberta, points out that when you type a prompt, the system translates the text into numerical encodings. For example, the model understands the data block and meaning behind a common word. However, it does not naturally focus on the individual letters making up that word. Because the algorithm processes blocks of meaning rather than strings of sequential characters, letter-by-letter breakdowns cause immediate confusion.
The takeaway for everyday search
This fundamental limitation has been a running joke in the tech community for years. Most famously illustrated by asking various chatbots how many “r”s are in the word “strawberry.” Sheridan Feucht, a PhD student studying LLM interpretability at Northeastern University, told TechCrunch that a perfect token system might not even exist, as chunking data is simply too useful for language models.
Bad spelling is not an urgent crisis for Google’s engineering teams. Still, it provides an excellent reality check for everyday web users. Currently, most advanced automation tools are far from all-knowing, and we cannot blindly trust them.
The post Why Google’s Powerful AI Still Can’t Spell the Word ‘Google’ Correctly appeared first on Android Headlines.
