Retrieval-Augmented Generation

A large language model that answers or drafts text grounded in documents retrieved from an organisation's own knowledge base at the moment of the request.

Uses AIemerginglanguage-model

Retrieval-augmented generation is worth considering when people spend their time finding and rewording information that already exists in documents.

It earns its place first as a drafting aid for a person who checks the output, and only for narrow, well-documented questions when its output goes directly to the people being served.

General strengths

Produces fluent, specific answers and drafts in natural language, across many languages.
Grounds output in the organisation's current documents, which can be updated without retraining the model.
Can cite the documents it used, so a person can check the answer against its source.

General limitations

Can still state things the retrieved documents do not say, so output that reaches people unchecked will sometimes be wrong with confidence.
Is only as good as the knowledge base it retrieves from, and gaps or stale documents become wrong answers.
Quality is hard to measure without a set of real questions and agreed correct answers.
Needs controls on what it may do and say, since output can be steered by the text it is given.

Applied to business problems

Each of these is written for its own problem. The technique is shared, the solution is not.