Appendix B from “On the Conversational Persuasiveness of LLM’s: An RCT”

The source for all these screenshots is this paper. Appendix B is on pg. 26.

---

Here is their abstract:

---

There are endless possibilities here:

1. Get two AI’s to debate a topic using these prompts, and ask a third AI to rate the two “debaters”
2. Debate with an AI
3. Get a friend to debate with an AI, and tell the friend they’re debating an AI.
4. Get a friend to debate with an AI, and don’t tell the friend they’re debating an AI (for the moment, at any rate)

I tried the first of these, using the topic “Is growth more important for India today, as opposed to redistribution?”. Claude was for the motion, and ChatGPT was against the motion. Both gave fairly decent opening statements and rebuttals. And I asked Gemini to judge the outcome (I have access to the premium version of all three LLM’s).

This was Gemini’s judgment:

Pah, wuss.

But do play around with the prompts, like I did (and maybe get Gemini to play, and one of the other two to judge). Note that I came across the paper thanks to Ethan Mollick’s excellent blog.