Why LLM’s can’t count the R’s in the word “Strawberry”

Understanding the Limitations of Large Language Models: Why They Struggle with Simple Counting Tasks

In recent discussions, a common point of curiosity has emerged around why Large Language Models (LLMs), such as GPT, often stumble when asked to perform straightforward tasks—like counting the number of times a particular letter appears in a word. For instance, model users frequently note that LLMs can’t accurately identify how many “R”s are in the word “Strawberry.” So, what’s behind this limitation?

The core reason lies in how these models process and represent language. LLMs work by segmenting input text into smaller units known as “tokens.” These tokens are then transformed into numerical representations called “vectors,” which are used throughout the model’s layers to generate predictions or responses.

Importantly, during this transformation process, the specific details of individual characters—like how many R’s are in a word—do not survive intact. Since LLMs are trained predominantly on statistical patterns of language rather than explicit character-by-character analysis, their internal representations do not encode precise letter counts. As a result, they lack the granular memory needed to perform tasks that require exact character counting.

For a visual explanation of this concept, see this detailed diagram: https://www.monarchwadia.com/pages/WhyLlmsCantCountLetters.html. Keep in mind, this article delves deeper into the technical aspects, making it a valuable resource for anyone interested in the inner workings and limitations of large language models.

Understanding these limitations helps set realistic expectations for what LLMs can and cannot do—and highlights the importance of specialized tools when precise, character-level tasks are required.