The task of finding relevant documents from a very large, unrestricted collection to answer questions, without being limited to a specific domain or dataset.
World knowledge accuracy, recall of facts and relationships
Performance retention over long documents and conversations