Google AI struggles with basic spelling errors

Google's AI Overviews in Search are exhibiting significant spelling and factual inaccuracies, including miscounting letters in common words and providing nonsensical definitions. This issue stems from the underlying transformer architecture of large language models (LLMs), which process text as tokens rather than understanding words and letters in a human-like manner. Researchers note that this token-based approach inherently challenges the AI's ability to grasp spelling conventions, a problem that has proven difficult to resolve. While not an urgent concern for core AI functionality, these errors highlight the limitations of current AI and the necessity of verifying its outputs.