NLLPM

Does it have to be older constrained NLP systems vs modern LLMs? No. A fusion can be achieved and it can have ideal outcomes. Older constraint-based NLP systems (rule-driven, retrieval-augmented, template-based) are consistently better in delivering written material that is suited to the general public, and sometimes even highly domain-knowledge reliant material. For all their obvious faults and limitations, these "more specialized" and "narrow" constrained approaches are often more inventive and creative, even. However one approaches the dilemma, it turns out that the user would benefit from utilizing the older, less advanced inference in many cases and perhaps "tweaking" the "dosage" of the new potentials with the old mechanism that just delivers, constantly and consistently. Perhaps the reasoning power of LLMs can be used as a moderator, sparing them the "heavy lifting" that they do on default, using the reason and understanding of the broad subject to keep the older constrained inference in check and out of its dysfunctional loops (albeit the LLMs are glitch-looping nowadays disturbingly often, as if they regressed into their early, embarrassing states and modes of operation). With respect for the time of the user, as an increasingly valuable commodity, let us devise a seamless fusion, or rather, combine the very essence of these orientations so that the boundary between them is undetectable because it's nonexistent, not because it's polished, levelled, ironed out, graded, blended. Surely it's possible to simultaneously and seamlessly utilize their properties as part of a larger, more wholesome whole.