At intervals of my choosing, I will resolve the answer asking a yes or no question about a word that is closest to 50%. Answers can also guess the word. Users are limited to one guess at a time (their first), and when I resolve a question I will resolve all the guesses as well.
Resolves YES if any user can guess the word. Resolves NO if nobody has guessed the word by April 29. Bear in mind that NO bettors on the main market can defect by intentionally asking shitty questions and betting in the wrong direction on individual questions.
There have been several previous games of this, such as this: https://manifold.markets/Stralor/manifold-plays-twenty-questions
Related questions
🏅 Top traders
# | Name | Total profit |
---|---|---|
1 | Ṁ77 | |
2 | Ṁ57 | |
3 | Ṁ35 | |
4 | Ṁ11 | |
5 | Ṁ10 |
An idea: you could resolve the question that's closest to 50% rather than a random question.
The question closest to 50% is the 'best one's in that it gives us the most information. And we can vote for which question is the best one by betting on them.
@Daniel_MC Given that we're doing this now, it makes sense to change the other rules as well.
In particular:
- Users can now submit as many questions (not guesses) as they want.
- Instead of mass resolving answers N/A, I will just resolve one answer at a time.
- I will mass resolve guesses every time I resolve a question.
Suddenly it makes sense why the other market linked in the description resolved the guesses closest to 0% and 100% - this lets users make a profit on their guesses if they're right. But given this market's not super active, you can probably do that anyway and still have me answer it.
@SaviorofPlant I like all of that. Was thinking a different threshold for a guess vs a question. Cause for a guess probably isn't a good one unless it's at like 90%.
You can't resolve a guess yes instantly if it's correct, because you're saying no to any guess that gets posted and you don't instantly answer
@Daniel_MC "Instantly" is a bit of an exaggeration, you're right. Edited it to "as soon as I see them"
@Daniel_MC People are limited to one question, I will ignore multiple guesses by the same person before the group of answers is resolved. That limits the number of guesses
What if you create less than 20 intervals before 29 April?
@KongoLandwalker Will resolve NO. I intend to do them every few hours on most days, so this will probably only happen if nobody is posting answers
@BrunoParga The idea of the game is to create answers asking questions (so people can bet on them) instead of posting them in the comments, which was my original idea
@SaviorofPlant actually it turns out to be negative-sum, unless fees from trading on N/Aed answers are refunded.
@SaviorofPlant I was just shitposting.
Also, to be honest, I don't really get what's going on with this market, but I kinda low-key despise this sort of navel-gazing market that resolves entirely within Manifold itself, so I don't even want to know much.
Notice that this is not criticism of you as the market creator - you're just following your incentives. My criticism is targeted at the Manifold team for setting up such dumb incentives in the first place.
With reasonable costs to venture guesses, the evolution of this type of market would be complex and (to me) rather interesting. There is less intrigue given the costs of guessing and the variability of windows between guesses, but in principle, there would be many interrelated bets at the resolution of each question.
@alieninmybeverage I can totally see how this is interesting to people. I just don't like that Manifold is so navel-gazing. It feels like a waste of potential.
I guess this site fell victim to one of the classic blunders, Goodhart's Law - Manifold optimizes for something bizarre and not entirely clear, but definitely not what I value.
@AnT I believe fees are refunded. When you resolve a YES/NO question as N/A, it says this: "Cancel all trades and return mana back to traders. You repay earned fees." I guess it's possible it's different for multiple choice questions though.
@BrunoParga In terms of mana, I am not really following my incentives here. The market is designed so that other people add the answers and get the mana reward for new traders. If we expand my incentives to getting attention, though, then I suppose this can be seen as goodharting.
@BrunoParga I think you are onto something generally, but I would also argue that there exists no "maximization" or "optimization" that does not have, as a consequence, some unintended target that then follows the same law. If anything, the best you can hope for is that the new targets offer some net balance or counterbalance. Which would be hard given a single currency and rule set for that currency.
@alieninmybeverage I don't think the principle you're stating is necessarily true; in very simple optimization spaces, we can probably craft counterexamples. Goodharting is generally something you'll see in high dimensional optimization problems, which could theoretically be avoided if you perfectly specified the optimization target. I'd love to discuss in more detail in Manifold DMs if you're interested.
@SaviorofPlant I agree that it's not necessarily true if we presume some idealization from the start. But the standard way of doing so is to idealize the variable for which one wishes to optimize or maximize. This begs the question as to the representative value of the measure, and hence becomes a target unto itself. If you want Mana to best represent positive predictive value in general, then you inhere the tradeoffs of both PPV and generalization, with some synthesized gains and some synthesized losses.
DM away if you'd like.