Google's AI has trouble with basic spelling, as demonstrated by its incorrect interpretations of words like 'Google' and 'journalism', according to a report from TechCrunch.
The Spelling Struggles
Google's AI Overview humorously suggests there are two 'P's in 'Google' and spells 'journalism' as 'j-o-u-r-n-a-d-i-s-m'. These errors are part of a broader issue as Google integrates generative AI into its search platform. Previously, the feature mistakenly cited satirical sources like The Onion, leading to advice such as using glue on pizza.
Google's Response
In response to the spelling inaccuracies, Google stated, "Counting within words has been a known challenge for LLMs, and we're working to fix this particular issue." Despite these efforts, spelling errors persist, while some other issues, like a misleading definition for 'disregard', have been fixed.
Understanding the Problem
AI researcher Matthew Guzdial explains that large language models (LLMs) like Google's AI are built on transformer architecture, which breaks text into tokens rather than understanding words as humans do. This process involves converting text into numerical representations, leading to errors in spelling. Sheridan Feucht, a PhD student at Northeastern University, notes that the inherent fuzziness in defining word boundaries for models makes perfect tokenization unlikely.
Implications and Future Outlook
While spelling is not the primary focus for researchers, these errors highlight the limitations of AI and the importance of verifying AI outputs. The utility of LLMs lies beyond spelling, but these issues remind us that AI is not infallible and should not be blindly trusted.
Source: https://techcrunch.com/2026/05/27/why-googles-ai-cant-spell-google-or-anything-else/
