How many Ps are in Google? According to Google, there are two.
Thereâs also is also âexactly 1 ârâ in the word âpoopâ,â Googleâs AI Overview says, as well as two âdâs in the word journalism, yet spelled it: j-o-u-r-n-a-d-i-s-m. Google did at least identify that there is one P in the last name of the U.S. president, but spelled it as t-r-p-u-m.
You didnât need to be a prophet to predict that Googleâs AI-forward Search overhaul was going to go over poorly. Weâve done this before. The first time Google added AI Overviews to Search, the feature ended up citing satirical posts from The Onion and Reddit, advising people to eat rocks and put glue on their pizza.
This time around, as Google doubles down on its commitment to make generative AI the centerpiece of its 29-year-old flagship product, itâs not surprising to see it stumble.
âCounting within words has been a known challenge for LLMs, and weâre working to fix this particular issue,â Google told TechCrunch in an emailed statement.
These basic spelling errors may seem familiar. LLMs, the kind of artificial intelligence that powers chatbots and other text-generators, are not built to understand spelling. Itâs been a running joke for years that whenever a company unveils a new AI model, you should ask it how many ârâs are in the word strawberry. These AI models â which can code an app in seconds, or solve problems that have stumped mathematicians for decades â are about as good as a kindergartener at spelling.
Googleâs AI overview woes reach beyond silly spelling mistakes though. Google already patched an issue from last week in which searching the word âdisregardâ would yield what looked like a dictionary definition of the word, only the definition was shown as, âUnderstood. Let me know whenever you have a new prompt or question!â But these spelling errors have remained amusing because theyâre so difficult to quash.
As researchers have previously explained when weâve asked about these spelling conundrums, AI doesnât perceive sentences as units of language made up of words and letters. Many LLMs are built on transformers models, which break down text into tokens, which can be full words, syllables, or letters, depending on the model. Instead of âreadingâ like a human would, the AI converts the text into numerical representations of itself, which are then contextualized to help the AI come up with a logical response.
âLLMs are based on this transformer architecture, which notably is not actually reading text. What happens when you input a prompt is that itâs translated into an encoding,â Matthew Guzdial, an AI researcher and assistant professor at the University of Alberta, told TechCrunch. âWhen it sees the word âthe,â it has this one encoding of what âtheâ means, but it does not know about âT,â âH,â âE.ââ
The token-based architecture that powers LLMs like Googleâs AI overview is inherently limiting, and researchers havenât been optimistic that they can solve the spelling problem.
âItâs kind of hard to get around the question of what exactly a âwordâ should be for a language model, and even if we got human experts to agree on a perfect token vocabulary, models would probably still find it useful to âchunkâ things even further,â Sheridan Feucht, a PhD student studying large language model interpretability at Northeastern University, told TechCrunch. âMy guess would be that thereâs no such thing as a perfect tokenizer due to this kind of fuzziness.â
This isnât necessarily an urgent problem on researchersâ minds, since the utility of LLMs doesnât come in their capacity to spell. But these blatant failures help us remember that AI is not perfect, even if it may sometimes seem like an all-knowing power beyond our comprehension. We cannot blindly trust AI outputs without double-checking their accuracy.
When you purchase through links in our articles, we may earn a small commission. This doesnât affect our editorial independence.
