The internet is buzzing with memes and screenshots of AI chatbots like ChatGPT and Claude misspelling the word “Strawberry”! They keep insisting there are only 2 R letters present in “Strawberry”, which is obviously incorrect.
In a time when advanced AI is tackling complex tasks like programming, data analysis, and solving intricate mathematical problems, it’s surprisingly prone to making basic mistakes, such as misspelling a word.
But have you wondered why is it that ChatGPT and Claude cannot answer this question correctly? Let’s understand why these cutting-edge tools can’t get to solve this simple question!
ChatGPT and Claude are “Berry Confused”
The whole “Strawberry” issue is quite interesting as you expect these Large Language Models to answer simple questions like these easily but here’s the catch. All of these generative AI chatbots are built from transformers. So, you need to know what they are and how they function to understand this entire problem.
How do transformers work?
These transformers are neural network architectures with various components that are used to train AI models.
The input text/prompt we give to the AI bot is converted into numerical representations. So, for instance, if you want to translate a sentence into French, the input will be transformed into numerical tokens.
Next, one part of the transformer tries to find out which words in that sentence are important and which ones are not. After understanding the relation of all words and figuring out what it means, it can process a language.
In short, it’s like a small team working out to know the words and how they relate to each other to effectively translate it. When it comes to “Strawberry”, the transformer again divides the word into inputs but it does not know the term contains the letters “S”, “T”, “R”, “A”, “W”, “B”, “E”, “R”, “R”, and “Y” in this sequence.
This makes it somewhat complicated for LLMs to count how many R’s are present in the actual word. However, the issue of fixing this spelling error is not easy to solve as AI chatbots like ChatGPT and Claude are built on transformers with the tokenisation function.
But here’s the good news. You can teach ChatGPT by clarifying the fact that the word “Strawberry” has 3 R letters and it will accept its mistake.
Gemini and Co-pilot answers correctly
To test out if other AI models can crack this, we tested this question with Google’s Gemini. Surprisingly, it correctly answered that “Strawberry” indeed has 3 R’s present in the word. The same response can be found by Microsoft’s co-pilot although it still apologises for making a mistake of incorrectly counting the letters.
The bottom line
AI chatbots like ChatGPT and Claude work with the help of transformers which tokenise the text prompts by splitting it. However, that does not mean these AI models understand individual alphabets, they only know the chunks of words. Perhaps, in the future, these large language models will be trained with different components to combat such problems but for now, they give us some hilarious memes!
