Retrieval-Augmented Generation / RAG
The method that checks and binds the data about an external source of information when producing the model.
Access-supported production (Retrieval-Augmented Generation, RAG) is a method that allows you to withdraw data from an external source of information (documents, databases, web pages) in real time and respond to the context of the model. The process is usually two-stage: the question of the user is converted into a embedding vector, and in a vector database, this vector has the most close document parts (retrieval); then these documents are added to the model prompt with the original question of the user, and the model produces the answer based on this binding (generation).
RAG’s major language models can be said that they are common for two basic weaknesses directly target: the training data of the model does not know the current events as it freezes on a certain date, and even if the model does not know, but it can produce false information (halusation). To produce answers based on documents, it increases the update of the answers and allows the model to be welded, therefore, verified. RAG has become one of the most common and cost effective ways of integrating its own private documents (inner documentation, product guides, legal texts) model into an artificial intelligence system without retraining; customer support boots have a wide range of application from corporate search systems.
RAG’s major language models can be said that they are common for two basic weaknesses directly target: the training data of the model does not know the current events as it freezes on a certain date, and even if the model does not know, but it can produce false information (halusation). To produce answers based on documents, it increases the update of the answers and allows the model to be welded, therefore, verified. RAG has become one of the most common and cost effective ways of integrating its own private documents (inner documentation, product guides, legal texts) model into an artificial intelligence system without retraining; customer support boots have a wide range of application from corporate search systems.