This guys gets it wrong too. It’s not even an engineering problem. It’s much worse: it’s a scientific problem. We don’t yet understand how the human brain operates or what human intelligence really is. There’s nothing to engineer as the basic specifications are not yet available on what needs to be built.

Will AGI require ‘consciousness’, another poorly understood concept? How are mammalian brain even wired up? The most advanced model is the Allen Institute’s Mesoscale Connectivity Atlas which is at best a low resolution static roadmap, not a dynamic description of how a brain operates in real time. And it describes a mouse brain, not a human brain which is far, far more complex, both in terms of number of parts, and architecture.

People are just finally starting to acknowledge LLMs are dead ends. The effort expended on them over the last five years could well prove a costly diversion along the road to AGI, which is likely still decades in the future.