Anagrams failure miscounting

I have ran into what seems to be a limitation of some sort. I asked the chat to make an anagram of Satoshi Nakamoto and the ts was not satifying at all. When I asked to begin the anagram in a certain way it began adding letters and miscounted the numbers of letters involved. Even when I specifically told the chat to use the correct number of letters it failed over and over. Anyone else found something similar?

Hi @ytgpkg9qkg :wave:
Welcome :people_hugging: to the community!

In general, LLMs are pattern-recognition systems, not symbolic calculators. When you try to count letters it may not count correctly because it understands tokens, not number of letters (for now). But LLMs have been improved.

You can use Python code alternatively.

Also, the number of characters does not equal everywhere same number of tokens.

1 Like