抄録
This study reports on analysis of feedback on learners’ reports from ChatGPT and their subsequent revisions. First, some of ChatGPT's "proposed corrections of unnatural parts" were inappropriate, but learners tended to choose whether to accept ChatGPT’s feedback based on their own judgement. Moreover, ChatGPT’s feedback differed in quality from that provided by the teacher, and it offered high praise even for parts the teacher did not comment on. While two learners revised their report in response to ChatGPT’s "questions that ask in depth about ideas", many of these questions were inappropriate and did not facilitate modifications to their reports. The results of this study suggest a need to reconsider both the prompt text and the use of feedback when using ChatGPT.
https://ror.org/03ynn1n59 
