I’m studying about, and testing ChatGPT. Since I’m studying Japanese, I thought about using it to help me in this task, too.
One area of Japanese I lack is output: talking or writing to other people. So I decided to try to make ChatGPT act as a chat companion, and also a teacher telling me how I can write in a more natural way.
However, I’m simply unable to make it do both things at the same time.
I try to be very explicit in my commands, telling ChatGPT to do both things. Sometimes it understands it for the first message, correcting my message and responding to it right after. But after that, it forgets to correct what I write and just responds to my message. Or it’s the opposite, it just corrects my message, but doesn’t continue the conversation.
My newest test was:
“From now on, for every sentence I write in japanese, I want you to do 2 things:
1. First, rewrite every sentence I write, suggesting a more natural way for me to say the same thing in japanese. Please explain why your suggestion sounds better.
2. After that, please respond to my message and continue the conversation.
Did you understand my instructions?”
Its response was:
“はい、あなたの指示を理解しました。以下、あなたが日本語で書いた文章に対して、リライトの提案とそれがなぜより自然な表現であるかを説明した上で、返信を続けます。 ”
It sounds like it understood my instructions. However, whenever I write something in Japanese in this chat, it corrects what I wrote and gives an explanation, but it doesn’t continue the conversation. If I remember it that it should continue the conversation, it continues the conversation but stops correcting my sentences. Does anybody know how (and if it’s possible) for ChatGPT to do both things?
3 comments
I’m sure someone was successful.
Unfortunately, as you can discover by searching ‘ChatGPT’ on this sub, they are vastly outnumbered by the people who got back convincing bullshit, unnecessary corrections, and bizarre tangents, all in impenetrably verbose paragraphs where single sentences would suffice.
Even the Japanese written here seems… a bit off. Stuff like リライト where I would expect 訂正 or 書き直し, or where it says それがなぜより自然なのか where the way it’s written is a bit confusing (なぜその方が自然なのか, or even just 訂正の理由 would be much simpler).
There’s a certain level of subjectivity involved that makes it harder to call some of these ‘objectively incorrect’, but a lot of ChatGPTs work just feels weird at times.
ChatGPT doesn’t “understand”. It’s literally just very fancy autocomplete and when it says that it understands something, that’s because that’s a likely way for that kind of conversation to go. It’s difficult to make instructions stick. It doesn’t weigh them differently from your other input. So they tend to work better at the start but as the chat progresses they’ll be displaced by all the other things you talk about.
Anyway for learning it’s not that good anyway, especially the older model that’s free to use. Maybe one step above using Midjourney as a picture dictionary but that’s about it. For output practice, get one of those language exchange apps like Tandem or HelloTalk, or you can pay someone to tutor you on iTalki.
Dogen just did a funny skit on this topic:
https://www.youtube.com/watch?v=DngDfh7MyeY