RAG – Retrieval Augmentation Generation

is a combination of Retrieval Augmentation and Generation to improve natural language processing tasks.

• Retrieval Augmentation: Use of text snippets that already exist in a large corpus.
• Generation: Use of neural models to produce text from scratch.
• Generation: Flexible, but prone to errors and inconsistencies.
• Generation: But improves answers because it can handle unseen or unexpected queries.
• RA + G: First retrieve the text that matters, then use a generative model to augment it.
• RA + G: The combination improves the coherence, consistency, and context of the response.
• Other areas to follow: text fusion, data augmentation, and text rewriting.
• Products: GPT-3 and similares