Large Language Models in Application - Mitchell Klusty

Abstract: Software agents are used to act on behalf of users and other higher-level programs to perform tasks autonomously. Rule-based and so-called narrow AI learning agents have been used for decades, often providing interfaces between independent entities (physical world, independent applications, etc.) as part of complex distributed systems. The advancement of multi-modal (text, image, signal, etc.) generative AI greatly expands the capabilities of autonomous agents in many domains. This talk will describe the use of software agents used in real-world applications. We will cover the integration of AI both through (using agents to support operations) and within (using AI with agents) complex agent-based systems. We will briefly describe the high-level process of custom LLM dataset creation and fine-tuning, a process that embeds information within the model itself. We will also cover the use of vector databases, which are used in Retrieval-augmented generation (RAG), a process that uses LLMs to access information that is not trained within the model. In addition, time will be spent on how agents make use of API specifications to explore and bridge information between external systems and interface with the physical world.