I'm running an inverse Turing test with two participants codenamed "AM" and "Multivac".
The AI is the free version of ChatGPT. ("GPT-3.5") The human is a volunteer from Manifold, who agrees to not bet in the market and not communicate with anyone about the test until it's over. (Their identity will be public. Feel free to volunteer in the comments and I'll pick someone I think can do a good job.) The participants are each randomly assigned one of the code names.
Anyone on Manifold can post a question, and I'll solicit answers from both the AI and the human and post those answers. Each user gets to ask one 1 question only. I'll resolve the market once it looks like no new questions will be forthcoming soon.
Since ChatGPT won't reliably pretend to be a human, the human's goal is to pretend to be ChatGPT. I'll send them a manalink for M$100 if they succeed in getting the market to be at at least 25%/75% in the wrong direction.
The human is allowed to use the internet to research their answers, but may not use the output of ChatGPT or any other AI, nor ask other humans for advice. (Including searching the internet for snippets of text from AIs. Not allowed.)
I won't bet in the market.
If you'd like to be the human in the test, please post a comment saying so.
Rules subject to change as needed to keep the game fun and fair.
1,000
3.00@IsaacKing Presumably the human needs a decent amount of familiarity with ChatGPT’s style to be realistic. Do they have to do that imitation totally from memory? Are they allowed to use ChatGPT while the market is open for things unrelated to this market?
@IsaacKing Since the market is open for a month, this might be a big ask from anyone interested enough in the project to want to be the human player