The issue here is one of semiotics and morphemology. Mapping meaning into a narrative and ontological protocol is going to be the requisite work if we want the

The issue here is one of semiotics and morphemology. Mapping meaning into a narrative and ontological protocol is going to be the requisite work if we want the engine to be "smart." As explored in the discussion at hand, tokenization creates a great mimic but it's a parlor trick. We must employ a robust thinking-thing that correlates not only a static, contextually indexed dictionary <lexicography>, we must also route that through a network to distill meaning itself into tokens. Perhaps languages which rely on morphemes for written language - a logosyllabary - are somewhat more or less suited for this task? I ask as a dummy. There also exists the consideration of allographemical contextualization, the nature of relevance, pragmatics, conjunct identification of context, semantics. To be honest the linguistics side alone is vast. Knowledge and cognition however. . . A whole other ballgame. But the only tool we have to really get down to the bottom of how knowledge works is language, it's to epistemological pursuit what math is to physics. While GPT is super impressive and can do a lot of quasi-brute-force things, we're only finding now the rudiments of the machined intelligence paradigm, and it will behoove any reader to brush up on their classics, true pursuants of philosophy and many order logic are about to be in high demand if I had to reckon.