I will give gpt-4 the prompt: "Generate some interesting yes or no questions for a forecasting competition." I will filter these for questions that already exist on manifold and for nonsensical responses, but will otherwise post them in the order that gpt gives them. I will post one question each morning before 9:00 EST.
This question resolves a month after I post the last question (February 11th) based on whether I make back the 250 mana required to post those questions. Not every question needs to be profitable; I just need a net profit.
Related questions
🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ466 | |
| 2 | Ṁ217 | |
| 3 | Ṁ157 | |
| 4 | Ṁ82 | |
| 5 | Ṁ80 |
@TheBayesian I have left myself reminders in early February and will be sure to remind anyone who reacts to this comment (with a like) if they so wish
A little curious about the source of profit here. You can only get back the 250 mana if every question gets 10 people to bet on it, and then you can only equal it. Are you counting some other way to make a profit (betting on the questions or something)? Or am I missing something?
@kalassak I've been receiving 5 extra mana for each new participant in the questions, so I will be judging this on whether I get an average of ten participants in the five questions.
i did this once, those turned out to be some very ... popular questions here. including /firstuserhere/will-an-ai-be-granted-legal-personh . Most of them were just asking gpt-4 to be creative, imaginative, innovative after giving it a few articles from quanta magazine and some ai related writing and telling it to dream lists of 20 specific questions on so and so
@firstuserhere Yeah, I've been thinking about the paper by Zhang et al about humans preferring AI writing (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4453958), and I wondered how GPT 4 would fare here. Even though this is not a rigorous investigation, I wanted something like a baseline before I tried any prompt engineering. Also given the questions I'm getting, it clearly has little grasp of realistic timeframes.
@Soli Likely so. Depending on the results I'll likely try some prompt engineering... but for now I'm curious what its baseline performance will be.
@AxelJacobsen Yes, I used the prompt in the description and am using its responses for the questions.