How many Ps are in Google? According to Google, there are two. There's also “exactly 1 ‘r’ in the word ‘poop’,” says Google’s AI Overview, but it spelled journalism with only two 'd's: j-o-u-r-n-a-d-i-s-m. Google made at least one correct spelling of the U.S. president’s last name—t-r-p-u-m—but that’s about as impressive as a kindergartener.

These basic spelling errors are not new to AI, and they’re certainly not limited to Google. Whenever a company introduces a new AI model, you can expect a joke about how many 'r's are in the word strawberry. These models, which can code an app or solve complex mathematical problems in seconds, are about as good at spelling as a kindergartener.

Google’s AI overview issues extend beyond silly spelling mistakes. Searching for the word “disregard” yielded a dictionary definition that looked suspiciously like, “Understood. Let me know whenever you have a new prompt or question!” But these errors remain amusing because they’re so difficult to fix. Researchers explain that AI doesn’t perceive sentences as units of language made up of words and letters. Instead, it converts text into numerical representations, which are then contextualized to generate responses.

The token-based architecture that powers LLMs like Google’s AI overview is inherently limiting. As Matthew Guzdial, an AI researcher from the University of Alberta, explains, ‘When you input a prompt, it’s translated into an encoding. When it sees the word “the,” it has this one encoding of what “the” means, but it does not know about “T,” “H,” “E.”’ Researchers haven’t been optimistic that they can solve the spelling problem. Sheridan Feucht, a PhD student studying large language model interpretability at Northeastern University, says, ‘There’s no such thing as a perfect tokenizer due to this kind of fuzziness.’

While these issues may seem trivial, they remind us that AI is not infallible. We cannot blindly trust AI outputs without double-checking their accuracy.