@mods I misresolved the market about the taliban. Could that be resolved false?

@NathanpmYoung done

@Bayesian How come i wasn't able to do this myself

@NathanpmYoung Only mods can unresolve options now I think

@Bayesian I am a mod.

@NathanpmYoung oh lol mb. you should be able to then. however:
idrk whether this button works:

and instead you gotta go on a specific option, press the ..., edit, and type unresolve to unresolve a specific option.

@NathanpmYoung Wow GPT is bad at this

@Yiddishe_Kupp1980 I mean yeah a bit. But doing it myself would have been stressful. What would have been better.

@NathanpmYoung You clearly stated that you were using GPT (and which model) so no complaints here about the resolution, but just surprised by this one. I think it shows the limits of these models.

Honestly ChatGPT seems more pro harris than I am.

Yes resolution counter to the market.

Surprising yes resolution. Pretty surreal to read

@NathanpmYoung so close and yet so far

It did not want to both google factchecks and end in the word true or false

Economic health resolves slightly counter to market.

Project 2025 resolves counter to the market

@NathanpmYoung I was confused by this statement- on the graph it looks like it ended a bit below 50% - until I checked the trade logs: someone bid it up to 75% from 45% around three minutes before closing. Almost literally a last-minute trade.

I think it's pretty likely the market would have corrected if there had been time. I'd probably have bid it back down, for one, I bought NO at a way worse price than that.

I wanted to do no interpretation at all, so this one didn't have a definitive true or false so I changed the prompt afterwards. Surprisingly this changed the valance of the answer. Perhaps if I did it again, I would use several models.

Before:

Afterwards:

I will post the surprising ones then.

Do you lot want me to post every single factcheck as a separate comment? That feels obnoxious, but will probably get more discussion. Currently there are going to a document here: https://docs.google.com/document/d/1yPrfRUqeHHUySYw59H0AzvGLmcZJFNJxAs98Gcp06ps/edit?usp=sharing

I prefer separate comments, since any of them might end up as discussions.

@ijk1 I think if people want to talk about one (or several of them together), they could just repost it here, right? We don't need to do that for every single one just in case.

@NathanpmYoung Friendly ping. ☺️ I know this is a bit of work to resolve -- just hoping to free up some liquidity.

@yetforever Yup it’s a high priority for me.

Two things I found useful for prompting ChatGPT:

• To stop it from just hedging and saying "it's complicated", it helps to ask "if you had to say this statement was true or false, which would you go for? Be clear about the rating you give."

• Explicitly asking "Use your web search feature if needed" helps ensure it always pulls context and sources from the web.