After transcribing the speech into text, it seems to first translate it into English before send it over the GPT-3 (davinci). I wonder if that will lead to a bu

After transcribing the speech into text, it seems to first translate it into English before send it over the GPT-3 (davinci). I wonder if that will lead to a bunch of mistakes, can't GPT-3 handle Mandarin directly? I did some experiments and using the same prompt but without translating it seem to lead to better results.