Retrieval-Augmented Generation combines retrieval of external material with a generative model, helping it use relevant information for a question. It can supply documents beyond the model's direct knowledge, but does not guarantee factual accuracy.
Retrieval and generation
Retrieval selects potentially useful material; generation organizes an answer from the question and available context. Wrong passages, overlooked limitations and incorrect synthesis can still produce errors.
Updating information is not retraining
Adding a file to a knowledge base generally does not change model parameters. This distinguishes RAG, current context, long-term memory and training, and clarifies an answer's sources.
Checking an answer
For a public manual, examine whether citations exist, support the particular conclusion and refer to the applicable version. A link is an invitation to inspect, not proof of a match.
Does RAG eliminate hallucinations? No. Retrieval, material quality and generation all influence the result.
Continue with What are AI models and large language models? to compare related ideas.
For a related project perspective, explore LingoAI’s AI work.