What is the rationale for superhuman intelligence? Neural networks are approximators being fed human intellect. Therefore they can only approximate the intelligence of humans. Even if the llm speaks an alien language, it should be similar to human intellect. Moving to the vertical axis would require some different mechanism.