The linked Arxiv paper also says humans could tell they were talking to AI 3/4 of the time. It would be interesting to do a follow up study to translate all hu

The linked Arxiv paper also says humans could tell they were talking to AI 3/4 of the time. It would be interesting to do a follow up study to translate all human responses to "sound like AI" in order to eliminate the possibility that humans are finding themselves more easily convinced by GPT-4 because it is GPT-4. Page 31 is a super interesting page too.