Neuroscientist: The bottleneck to AGI isn’t the architecture. It’s the reward functions: a small set of innate drives that evolution wired to learned features of our world model, and that gives rise to generalization.