RAG

Spread the love

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

Leave a Reply

Your email address will not be published. Required fields are marked *