Thanks again for a super thoughtful response and for the offer to help. How would you prefer to collaborate going forward? I am happy to do so here in LW comments in case perhaps others find it helpful but would also be happy to do so on Reddit ,offline, etc. until we think we have something worth publishing—good point on half baked prepper ideas potentially being net negative.
people who have already identified a scenario that they find compelling
That’s a super good point, my and my network’s expertise lies more in bio risks (and to the degree these overlap with AI scenarios). Perhaps it is better to start with bio—it also seems easier to tackle as many people seem to think AI is kind of binary (as in either everyone dies or most people survive) and also much harder to prepare for. Then, if that somehow becomes successful, we can later on consider venturing out with advice for prepping for other scenarios (e.g. if AI destroys most democracies and people could need plans for escaping to the last liberal democracies standing or building a new one).
Maybe someone else commented something similar here, but I encountered this issue for the first time in an internship in a bank in 2007 where we had a fake investment competition. An option trader told me that a competition is really a binary option, with 0 value if less than strike price and a big reward if above. This means you want to maximize the value of the option which in this case means maximizing volatility. I crushed the portfolio competition by betting on leveraged financial instruments correlated to oil price (the biggest volatility investment at the time).
Another issue I have not explored with GJopen is the possibility of adjusting for past over or under estimates. For example, if you for some period in the beginning forecasted 70% probability of A and 30% of B and then mid way through the question get new data indicating that B is actually 70% likely, you might not want to put 70% but some higher percentage to compensate for the period before having a too low probability. So you can artificially lower your Brier score due to the forecasts being made over time. I am sure there are ways to account for that, I just have not had time to think about it.