Hacker News new | ask | show | jobs
by ssivark 2201 days ago
That may or may not be the case, but note that there are many (most?) domains where the positive payoff is bounded, but the negative payoff is not. This method seems particularly useful for those scenarios -- if you have "adequate" performers to copy from.
1 comments

Instrument flying is actually a great example of this. There are no instrument flying competitions. Well, I guess there’s one every flight. First place, you make it there. Second place, you diverted. Third place, you crash and probably die.