It's still SO early. We are in the "640K [of memory] ought to be enough for anybody" phase of LLMs. So much more to go.