My intuition is that something like this (beyond simple transformer architecture) is necessary to reach real AGI/ASI. The good news is that these additional layers might be relatively easy to build. A small fraction of the resources (both compute and human capital) devoted to hyperscaling might get us there. I see a lot of innovative ideas like this coming from the Sinosphere. IF hyperscaling isn't the path to AGI/ASI (as opposed to clever new ideas and better architectures) then the US lead over China might be zero or even negative!

@brandonwilson ここで指摘したように、世界について考える真面目な人は、特にAIやテクノロジーにおける米中競争を理解しようと努めるべきです。