@lucashendren I'm an outsider, so I don't know that I have insight. I just created this map as a way to visualize some places where something could emerge. But even with transformers we're seeing coupling them with scaffolding for long horizon memory and structured schema (via knowledge graphs) in highly verifiable domains is unlocking performance gains. I imagine we're still exploring how useful that can be. There are also clearly many narrow use cases for transformers.