Introduction
Large Language Models (LLMs) have revolutionized the field of natural language processing, enabling machines to understand and generate human-like text with remarkable accuracy. However, despite their impressive capabilities, LLMs are not immune to limitations. In this article, we will explore one such limitation that has sparked debate among language model enthusiasts: their inability to accurately count the number of specific letters in a given word.
Understanding the Issue
Some people still post about the LLMs’ inability to tell how many particular letters are in a given word. Let’s take a look and try to understand the basic issue here.
This shortcoming comes from tokenization. Large language models don’t see text as letters, they see it as tokens. A token can correspond to a single word, a syllable, a phrase, or a single character.
Seeing is believing, so take a look at how [Llama 3](https://belladoreai.github.io/llama3-tokenizer-js/example-demo/build/), [Anthropic](https://lunary.ai/anthropic-tokenizer), and [OpenAI](https://platform.openai.com/tokenizer) tokenize texts.
Llama 3:
OpenAI:
For an analogy, imagine asking how many letters E are in the word 77. This question makes no sense. Only if you expand “77” into “seventy seven” you can see that there are four letters E. LLMs can’t do that.
What Can LLMs Do
One way an LLM would be able to answer the how many letters questions is if it saw a sentence saying “the word XYZ has three letters A”, either during training, or maybe in text pulled with RAG at the answering time.
Also, if a model has access to tools, for example a Python interpreter, it can figure it out by writing code. For example, claude-3.5-sonnet in Cursor provides the code without being asked to do so.
To determine how many times the letter ‘a’ appears in the word “abracadabra”, we can use a simple Python script:
word = "abracadabra"
count = word.count('a')
print(f"The letter 'a' appears {count} times in '{word}'."If you want to give a model a fair chance with the letters, separate them with spaces so that each letter is a token.
Another answerable question would be, how many tokens are in the word “strawberry”, as an LLM should be able to answer this. However, that is should doesn’t mean that it will.
You can try asking a model something like “in your internal representation as an LLM, how many tokens does the word XYZ consist of?”
This could potentially provide clues about models of unknown identity. For example, the perplexity.com chatbot says the word “strawberry” is primarily represented as a single token in my vocabulary, which would suggest that it’s Llama. However, it also says the word “abracadabra” is a single token, which is unlikely.
Conclusion
Large Language Models have many impressive capabilities, but their inability to accurately count specific letters in a given word is a limitation that cannot be ignored. By understanding the underlying issue of tokenization and exploring alternative methods, we can better appreciate the strengths and weaknesses of these powerful language models.
Frequently Asked Questions
Question 1: Why do LLMs struggle to count specific letters?
LLMs struggle to count specific letters because they see text as tokens, not individual letters. Tokens can correspond to single words, syllables, phrases, or single characters, making it difficult for LLMs to accurately count specific letters.
Question 2: How can LLMs overcome this limitation?
LLMs can overcome this limitation by seeing a sentence saying “the word XYZ has three letters A”, either during training or at answering time. Alternatively, LLMs can write code using a Python interpreter to figure out the answer.
Question 3: Can LLMs count tokens in a given word?
Yes, LLMs can count tokens in a given word. For example, an LLM can say that the word “strawberry” consists of a single token.
Question 4: Can LLMs answer questions about letter frequency in a given word?
Yes, LLMs can answer questions about letter frequency in a given word. For example, an LLM can say that the letter “a” appears four times in the word “abracadabra”.
Question 5: What are the implications of this limitation for LLMs?
The implications of this limitation for LLMs are significant. While LLMs are incredibly powerful language models, their inability to accurately count specific letters in a given word highlights the importance of understanding the underlying mechanisms of language processing and the limitations of these models.
