GDIT Exec Sees Benefits of Adaptive RAG in DOD GenAI Use

Dave Vennergrund, the vice president of artificial intelligence and data insights at General Dynamics Information Technology, recommends that the Department of Defense and its partners enhance their retrieval augmented generation deployments of generative AI with adaptive approaches.

Conventional RAG

Conventional RAG deployments work to improve the relevance of GenAI outputs and reduce hallucinations by allowing users to augment queries with additional sources of information or domain-specific data to produce more specific responses, Vennergrund explained in a recent column on the GDIT website.

Adaptive RAG

But the answer provided by GenAI running on conventional RAG will “only be as good as the prompt,” Vennergrund pointed out. Adaptive RAG addresses this deficiency by modifying the prompt and running the query again if the initial response does not meet a standard of quality or relevance threshold. The assessment of the initial response can be done by the adaptive RAG system itself if it features a self-reflection mechanism.

AI Trustworthiness

Vennergrund sees tremendous potential for adaptive RAG in defense use cases, especially in light of increasing demand for GenAI across the DOD and the need to ensure the trustworthiness of the technology’s output. The GDIT executive also believes deploying adaptive RAG “will allow DOD mission partners to drive efficiency and improve result quality.”