There is still innovation in the space. There are some models that aren't as gpu (and maybe power) hungry.

I think we need to be prepared to confront LLM's and whatever comes after for the long haul.

Today's iteration sucks for so many reasons, the core of problem I think will be there for a long time.

Hell we may crack fusion (in the next 20 years) and we will still have the fundamental problems we are grappling with.