@samschapiro "Empirically, we find...." - I am wondering if this is empirical research. Let's say I study the correlation between rain and temperature, but only use weather forecasts, no actual measures. I wouldn't call that empirical.
@RobinNa_@Kehang_Zhu@locross@bakkermichiel Nice OOD design. But, grounding is double-edged. In-family profiles help; out of family they can do worse than nothing (−68). But the paper defines the families up front. In practice you'd need to know a priori where OOD starts, and that's the hard part.
@felixrieseberg@GergelyOrosz Cloud makes sense for some use cases, but removing options looks shady. I have powerful computers on the edge, and I'd rather have a discount for not using your cloud vms.
@johnjhorton@BenSManning Simulation may benefit from individual agents, but it is also where their small errors compound instead of averaging out. And simulation isn't the purpose anyway. The decision is.