Artificial Intelligence, especially LLM technologies, are quickly evolving to find new applications and more effective techniques.

A revolution that started with the transformers for neural nets birthed the likes of ChatGPT and similar LLMs. This in turn has opened the flood gates to new concepts, ideas, and techniques.  As LLMs flood the directly applicable domains, CSI researchers (and others) are hard at work developing new variations and enhancements to address weaknesses or expand capabilities for domains where traditional LLM strategies fall short.  Newer variations include Retrieval-Augmented Generation (RAG), integration with Reactive and Cognitive Agents, fusion with prior forms of feedback learning and Bayesian and probabilistic techniques, and many others.  In each case, new LLM hybrids have new capabilities and can address new problem spaces where pure LLMs proved inadequate.

CSI has integrated LLM support for natural language query, tasking, and decision response, in addition to unstructured/semi-structured data transformation and query result transformation to NL form. Using the LLM NLP translation, the formal semantics foundation, and the automated planning over a rich, semantically grounded plan representation provides a context, history, and situational foundation from which to automatically build, integrate and manage the domain concepts.  Using the domain of operations and logistics orders, planning, and plan execution provides a structure in which to explore the concepts, build capability leveraging extensive existing work, and evaluate the effectiveness of the new approaches.