So many of these articles jump around to incredibly different concerns or research questions. This one raises plenty of important questions but threads a narrat

So many of these articles jump around to incredibly different concerns or research questions. This one raises plenty of important questions but threads a narrative through them that attempts to lump it all as a single issue. Just to start off with, saying LLM models are "not smart" and "don't/won't/can't understand" ... That is really not a useful way to begin any conversation about this. To "understand" is itself a word without, in this context, any useful definition that would allow evaluation of models against it. It's this imprecision that is at the root of so much hand wringing and frustration by everyone.