My chat GPT is making many mistakes such as I asked it to say me a word with 4 letters and It answered me a word with 6 letters. And this mistakes appear every time!
Sorry mine seems to work well. Must be the way you ask it. Use the word “see” instead of “say” .
I asked for one word of 4 and another of 6 and worked perfectly.
1 Like
Okay, thank you! 
But I have it in Spanish because I am Spanish, and I asked it to say me one word that is “aguda” and another that is “llana” and it told me 2 words that are “llanas”.
Hi @inesgol19 
Welcome 
to the community!
AI counts TOKENS. It does not count letters correctly every time. If you get correct answer once, it does not mean “Woow BINGO!”
Also, if you ask short words it may give good result, but with long words not.
I think you have already heard about “How many 'r’s in ‘strawberry’?”
Especially, it is hard for Non-phonetic language.
In Non-phonetic language, some letters may be silent, or the pronunciation of letters varies depending on the word.
It is easier but there is no guarantee in Phonetic language.
In Phonetic language, letters and sounds correspond directly, there is no lost sound.
You will see in image 1, answer is wrong, but if you ask in another chat you may get correct answer. It is possible.
You will see in image 2:
It is correct, but in another try it says there are two ‘k’.
You will see in image 3:
1st answer is FALSE
2nd answer TRUE
3rd answer is FALSE
3 Likes
Ok thanks! 
I will be carefully!!,
You’re right! @gdfrza
It’s not just Python in IDE or on ChatGPT, using any coding language, for example Apex, we can achieve great results counting.
However, most people don’t know how to code and prefer to use natural language when interacting with AI.
For example:
-
In the first image below, I asked the AI a question, and it didn’t interpret the capital “I” as an ‘i’. Instead, it seems to perceive it as something like the “Ayyy” sound. Interestingly, the AI says there are two 'i’s, even though the “ea” sound in both words sounds like ‘i’ to us.
-
In the second question, I asked a question in Turkish. Even though the sentence contains no “e”, the AI still claims there are two. This seems to happen because the letter ‘a’ in Turkish has a sound that resembles the English ‘e’.
The Turkish sentence I used is pronounced exactly as follows:
“Ayran içtim” which translates to “I drank ayran.”
Pronunciation:
Ay = I
ran = run
iç = each
tim = team
Interestingly, when I switched to Turkish, the AI provided accurate responses:
- In the first question, I asked about the letter ‘i’, and it correctly stated that there are 2 'i’s.
- In the second question, I asked about the letter ‘e’, and it correctly stated that there were no 'e’s.
Well done!
In Turkish, ‘a’ is always ‘a’, ‘e’ is always ‘e’, and ‘i’ is always ‘i’. Even if we combine letters like ‘ae’, ‘sh’, or ‘sch’, their sounds remain unchanged, unlike in English where phonetics can vary. For example, in English, the letter ‘c’ sounds like ‘see’, but in the word ‘school’, it sounds like ‘k’. Turkish doesn’t have these inconsistencies, which makes learning phonetic languages straightforward. For instance, if you speak Turkish, you could learn to read Swahili in a single day.
1 Like
Indeed. Also should be more ways to do kind of things and on future even more ways . For example some time ago you couldn’t play well tic tac toe against chatgpt and now you can using Lua and now actually knows when one wins or lose , you can try by yourself.
Should look like something like this:
And to get a interesting answer while counting letters, " customization" …
2 Likes
True, I just took a test with Japanese.
It’s more difficult in this language for several reasons.
To put it simply, it’s a syllabic script. A katakana or hiragana character usually consists of two letters, but not always.
Well, here is my example:
I asked for words with 4 and 8 letters.
CustomGPT’s answer:
“1. 月 (Tsuki - moon) - 1 kanji, but 2 sounds if we look at the syllables!
2. コンピュータ (Konpyūta - computer) - 8 katakana characters!
Japanese of course adds a whole new dimension to our little puzzle with the characters and syllables."
Here’s what happens:
- Tsuki has 2 syllables, “tsu” and “ki”. The “u” is practically soundless. GPT does not count the “u” as the 5th letter.
- Konpyūta consists of “ko”, “n”, “pyū”, “ta”, i.e. 8 letters.
I will try that with Japanese 
1 Like
Recently, ChatGPT kept saying “Year of the Dragon” when it should be “Year of the Snake.” Also, it’s “Chinese New Year”, not “Lunar New Year.” No matter how many countries celebrate it, we are all celebrating “Chinese New Year”—let’s not allow cultural appropriation!