--
You received this message because you are subscribed to the Google Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to everything-li...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/everything-list/CAJPayv1zBfaRbTmBgBbTMnFbhaEEU8U9RzuqXi6s4pzkxP-Ptg%40mail.gmail.com.
> I suspect that the fact that the participants could in most cases tell they were arguing with an AI is very significant.
John K Clark See what's on my new list at Extropolis
An article in the May 19 2025 issue of the journal Nature reports that if GPT-4 knows just a few facts about the person it is talking to then it is 64% better than a human with similar information at convincing another person that their conspiracy theory is wrong, and to do that the AI had to use more than just logic because the conspiracy theory was not originally caused by logic.
"The results showed that when neither debater — human nor AI — had access to background information on their opponent, GPT-4 and people were about equally persuasive. But if the basic demographic information from the initial surveys was given to the opponents before the debate, GPT-4 out-argued humans 64% of the time. When provided with even just this very minimal information, GPT-4 was significantly more persuasive than humans. It was quite simple stuff that normally can also be found online in social-media profiles.”
mpt
On Wed, May 21, 2025 at 2:33 PM Brent Meeker <meeke...@gmail.com> wrote:
> I suspect that the fact that the participants could in most cases tell they were arguing with an AI is very significant.
The only way they could tell is by noting that they were talking with something that was smarter, more knowledgeable and had a much much quicker mind than any human being.