Ms. Swathi Suddala

Advancing Generative AI with Retrieval-Augmented Generation Systems and Large Language Models (LLMs)

Abstract:

Legible language models (LLMs) are poised to revolutionize programming, feedback
generation, scripting, and automated testing systems. Despite their potential, recent
critiques regarding accuracy and precision have underscored several near-term
limitations of generative AI frameworks. To develop a first-generation production version
of retrieval-augmented generation (RAG) systems that enhance information access and
writing capabilities, we anticipated improvements in robustness, accuracy, and overall
performance of today’s leading LLMs, alongside the rapid deployment of API
integrations, user interfaces, and workflows.
As major data centers advanced hardware capabilities and infrastructure clouds scaled
software systems, new approaches to improving near-term capabilities emerged. These
include ultra-large language models featuring probabilistic reasoning and factored
representations, which are being introduced at this workshop. These advancements lay
the groundwork for enhancing RAG systems. Concurrently, ongoing research and
development span hardware, software, and neural model design and training, enabling
the integration of advanced features into hybrid cloud production AI systems.