I'm trying to use pol.is to workshop AI alignment questions: https://pol.is/4fdjudd23d
Background: John Wentsworth complained that we keep asking AI timelines questions instead of more interesting questions that would actually help him make decisions as a researcher. Of course, anyone can ask anything on manifold, but it's often hard to go from an uncertainty "what should I do next as as a researcher" to concrete questions that retain what you care about.
🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ62 | |
| 2 | Ṁ30 | |
| 3 | Ṁ16 | |
| 4 | Ṁ5 |
I think I would've used pol.is differently: instead of prompting people with "is this question useful?" and then picking the most upvoted (I assume this is what you intend to do), I would ask people to answer the question itself and then I would pick the most controversial ones (because those are higher value of information). Maybe I'm missing something / misunderstanding your intent. Also, not sure this makes a huge difference in practice.
