I think the future of LLM's if the funding winter does not come is going to be specialized low size models with specialized language(non-english) which can be integrated in small hardwares and can run on small compute.
Yapping at my "aha" momemnts
Hosted on streams.place.