Google AI fails basic spelling and counting tasks

google's rollout of ai-generated search overviews is facing criticism for failing at basic spelling and character counting, such as claiming there are two 'd's in 'journalism' or misspelling the word entirely. experts explain that transformer-based llms do not process text letter-by-letter but instead convert tokens into numerical encodings, making granular spelling tasks inherently difficult. while google frames this as a specific issue under review, researchers suggest the tokenization process creates a fundamental barrier to solving these counting errors reliably. these incidents underscore the limitations of current ai architectures, reminding users that llms require verification despite their advanced capabilities in coding or reasoning.