The Apple lab has a video demonstrating the difference between expressive and functional:

https://machinelearning.apple.com/research/elegnt-expressive...

I think this is the goal of LeLamp.