Also, there is a third kind of proposal that I think someone should try at some point (this is not my original idea, though this specific formulation is): basically, use forking into multiple chains to solve every problem. The idea is that if there are N decisions that have been made, then you have 3^N token types, where each token type is a length-N ternary string (eg. 012010221). When the system starts, everyone has only tokens of type 000000… Anyone has the ability to split tokens of type x0y, where x and y are arbitrary strings, into 1 token of type x1y and 1 token of type x2y (eg. 012010221 can split into 012110221 + 012210221), and you can also recombine.
Anyone has the ability to burn 1 coin of any type to ask a question; question IDs are sequentially assigned globally. At that point, anyone who has coins of that type or a subtype (eg. if the type is 012010221, then with the new assigned question ID it becomes 0120102210, and 1120102210 is a subtype) is expected to convert them into *1 + *2 pairs (eg. someone with 1120102210 would convert to 1120102211 + 1120102212), and sell the coin that they think corresponds to the incorrect answer and buy the coin that they think corresponds to the correct answer (this could be done on a DEX, or the 1 coin could be used to subsidize an on-chain market maker).
The goal here is that coin types that correspond to correct answers are the coin types that people would want to burn in order to ask the holders of those coins questions, as people feel those holders are more likely to give more correct answers, and so coin types that correspond to correct answers will be more valuable, and most other coin types will fall to such a low value that most of the time people will not even bother to send a single transaction of them (this is why this scheme won’t actually have exponential overhead in practice even if it seems like it might in theory).
You could also organize the scheme as a tree, where you can burn 1 coin of some type and at that point generate two child types that everyone must choose between, but the 3^N hypercube solution (yay, hypercubes!) is nice because it potentially allows decisions to be made in parallel, and it allows users to just not participate in votes that they are not sure about.