Problems with Korean prompts

[chrislemke]

Problems with Korean prompts

[chrislemke]

Hey @circumeo!
Hm 🤔, that is actually an interesting problem. Do you enter the prompt in Hangul or in the Latin alphabet?

[circumeo]

"비빔밥 레시피를 알려줘"
Thank you for the test! 😊

[circumeo]

It seems to be connected that ChatGPT always receives some parts of the prior conversation. In the workflow's configuration, there is the ChatGPT history length for adjusting this. You could decrease this number so the model only receives the n-newest requests. Alternatively, you could also use the forget me prompt which removes all the conversations from the cache. I will think of a way to improve this and may provide an update in the future. But for now, this seems to be the solution.

Thank you so much for explaining in detail! 😊
It helped me a lot to solve the problem :)

[circumeo]

Hey @circumeo!
Hm 🤔, that is actually an interesting problem. Do you enter the prompt in Hangul or in the Latin alphabet?

I have tested both cases.

If you write a prompt in English, you will print a normal answer, but if you write in Korean, you will come up with a wrong answer that doesn't fit the question at all.

For example, if you ask a bibimbap recipe in Korean, the story of building fashion is printed as the answer.
However, if you access the chat app site and ask the same question in Korean, the normal answer will be printed 😥

[chrislemke]

Could you provide an example prompt so I can test it?

[circumeo]

Could you provide an example prompt so I can test it?

Should I upload a screenshot or should I write down the sentence I asked?

[chrislemke]

I would prefer a sentence. So I can copy it.

[chrislemke]

It seems to be connected that ChatGPT always receives some parts of the prior conversation. In the workflow's configuration, there is the ChatGPT history length for adjusting this. You could decrease this number so the model only receives the n-newest requests. Alternatively, you could also use the forget me prompt which removes all the conversations from the cache. I will think of a way to improve this and may provide an update in the future. But for now, this seems to be the solution.