Tuning a model specifically to find and rank relevant documents or passages in response to a query, rather than generating new text.
World knowledge accuracy, recall of facts and relationships