🏅 Top traders

@Joshua resolves NO

Will this resolve 1 day before the other market closes if it is at or above 50?

@Joshua

@TobyBW Oh, great point. Yeah this can resolve early if a yes resolution becomes impossible

that’s a big drop based on a very transient event.

Why are people holding YES here? Other markets show ~10% for a NO in Mira's market, and we already have a successful solve, so I don't understand why that would increase.

Has anybody posted a partial resolution already?

In that case, I guess @Mira’s market is irrational when it is below 50%. Right?

@IonMarqvardsen Nobody has yet solved a single freshly-generated puzzle, in a way that replicates, much less either 20% or 80%.

There have been some solves, but they either took too many turns so weren't eligible, or didn't replicate during testing. So the official count remains at 0.

@Mira Then I don't really understand why the market is so high, despite so many unsuccessful attempts. Any idea? Is it because people are not aware of the replication failures and are still convinced that partial resolution has been achieved?

@Grinchtachu Many people were waiting for OpenAI DevDay on November 6 for the price to decrease to really step up their testing. So in December, I'm expecting there to be one last push to the finish line using the new GPT-4 Turbo model.

Before, it was easy for people to think somebody else would solve it, or maybe there would be a price reduction. Those possibilities are getting more unlikely, so people that have been waiting will now attempt it in earnest.

You can also bet on my attempt. Since I am judging the contest, I didn't want to submit a solution, but if nobody solves it I will give it a try myself:

/Mira/will-mira-solve-the-2023-sudoku-cha

@Mira The hopes from price decrease and GPT-4 turbo should be behind us now. It is even possible that GPT-4 turbo is less suited to resolution attempts than prior version. According to discussions on the discord, now it sometime refuses to comply with a request when it considers the number of steps would be too high.
It is always possible that someone is hiding some progress, but I doubt it.

@Zardoru

It is even possible that GPT-4 turbo is less suited to resolution attempts than prior version.

It suits just fine

Displayed probability, right? And days are defined midnight-to-midnight?

@Mira Displayed yes, but I had thought any continuous 24 hours should count.

@Joshua There was a 29-hour gap between November 24 10:44 and November 25 15:43 if you look at the underlying value:

However, I don't see any 24-hour gaps yet using a threshold of 49.5%, which displays as 50%:

Using displayed probability is the convention for these, because you could have a YES limit order at 50% and still the probability might be 0.4999999997 or something.

But since one interpretation would already resolve YES, it seems good to make this convention explicit...

I'm fine with either interpretation, too.

@Agh I was buying it up when it showed less than 50% in the UI and thought there were no 24 hour gaps. But I had to check in more detail just now because I saw you buying a lot of YES shares against me...

@Mira Yeah 49.5% doesn't count if that displays as 50%, Description has been edited.

Why would the market close above 50 if it's going to resolve NO? Shouldn't it trend towards 0 as market close approaches?

@DavidSpies It resolves to 50 for a partial solution (solves 20% of sudokus rather than 80%, five attempts), so this market is primarily gauging whether the recently submitted solution at least hits that 20% mark.

@DavidSpies There could also be a savvy submitter who holds back a successful prompt until the last minute, but that seems unlikely.

@DanPowell I'm haven't looked hard into it but I am starting to worry this is the case.

Update: I made a market for this