“David Autor of MIT believes that simply scaling up LLM capabilities will not lead to AGI because it does not substitute for missing cognitive components.”