I'm a total noob in this area of research but is trying to jam all human knowledge into these models the correct approach? Would a lightweight multimodal model

I'm a total noob in this area of research but is trying to jam all human knowledge into these models the correct approach? Would a lightweight multimodal model integrated with a knowledge engine like Wolfram alpha or databases of source material be a better approach? I'm optimistic. The amount of attention this area is getting is insane.