I think we need to start thinking of these networks not as something we hope will someday simulate a brain, but as small building blocks that can be networked t

I think we need to start thinking of these networks not as something we hope will someday simulate a brain, but as small building blocks that can be networked together in a variety of ways to tackle tasks. DALL-E may not understand 3D image manipulation, but if you chained the output from DALL-E into something like NeRF you could start to transform the image in 3D space. One can picture a collection of a few hundred of these networks, all with different things that they are good at working in concert to approach something that is very good at doing things generally. We're missing the layer on top that decides which algorithms to run and in what order.