Hi, so basically it’s exactly what I have written in the title. ChatGPT seems to have trouble in following character limits (ex: less than 300 words/300자 미만으로 해줘) in Korean. I’ve tried asking it in English and whatever English text it spit outs, it does a good job in taking that character limit into account. But when I ask the same direction in Korean for Korean texts, it absolutely does a horrible job.
Example :
Input : 자기소개를 해보세요 (100자).
Output : 넵 지시에 맞춰서 자기소개를 하겠습니다 (글자수 98자). 실제로 공백수 혹은 공백수를 제외한 글자수를 세보면 56자 혹은 지나치게 높은 120자 정도로 정확한 글자수가 맞지 않음.
It never gets the exact word count it was asked to do.
Hi, welcome to the community!
ChatGPT isn’t good at counting characters or words exactly no matter what language you use, including English or Korean. That’s because it doesn’t actually “see” the text the way a person or a regular computer program does.
Instead of reading each letter one by one, ChatGPT works by guessing what comes next based on patterns. It breaks sentences into pieces called tokens, and these tokens don’t always match words or characters exactly. For example, one token might be part of a word, a whole word, or even punctuation. This makes it very hard for it to count letters or characters correctly.
So when you say “write exactly 100 characters,” ChatGPT might try to guess, but it doesn’t really know how many characters it’s using unless it runs actual code, like Python, to count them.
For example, the sample below it says it wrote 318 characters which is wrong. But when it count using PYTHON code, it says 257 characters which is correct.
You can ask to trim it after writing. You can see sample below, it removes last two words after using python code.:
