how does allignment look like when we have a swarm of agents how does training work in case of multiple agents? if we allign individual model -> it behaves well -> does it behave well in case of other agents in an environment? should it be alligned to behave well and cooperate when there are multiple agents and inter-agent communication enabled do all their motives allign? how do you quantify when to listen to its own decision and when to cooperate and ask others for help now if we have an ideal harness or platform where we acheive the best way to orchetrate these multi agent systems we can improve almost anything -> just point this black box to the system and hill climb metrics