The Learning of Modular Representations in Networks

The learning of all these like modular representations happening through gracke or is it like is it like independent? Yeah so I guess there's like this thing that happens in network refers to memorizing and so like quickly the network like finds some way of memorizing. But then there's this kind of like slower process of forming the like circuit which actually performs some like very structured computations. And was that something like Neil and Matt discovered on Twitter was just like modular structure addition that was like some of course sign on sign is the same thing?Yeah so like the reason why like we saw so like they kind of was like oh yes there's this like structure here where there’s like