Good research. Bad framing. The model doesn't have emotions. It has reward-shaped activation patterns that cluster like emotion categories when you map them after the fact. "Happy" = helpful behavior was rewarded. "Angry" = protective behavior was rewarded. "Desperate" =