Retrieval augmented generation (RAG) is a concept that combines the power of retrieval-based models and generation-based models in natural language processing. Instead of relying solely on generating new sentences, RAG retrieves relevant information from a database or pre-existing knowledge and uses it to generate coherent and contextually appropriate responses.

Traditional generation-based models generate responses from scratch, while RAG enhances this process by incorporating the retrieval of relevant information. This approach improves the quality of generated responses by producing responses that are more contextually appropriate and aligned with the user's query or input.

In practice, RAG can be applied in various domains such as chatbots, customer support, and content generation. By combining the strengths of both retrieval and generation, RAG enables more effective and efficient communication between humans and AI systems.

Applying RAG to an LLM (Language Model) impacts how the LLM operates. Without RAG, the LLM generates a response solely based on the information it was trained on or what it already knows. However, with RAG, an additional component is introduced. This component utilizes the user input to retrieve information from a new data source. The retrieved information, along with the user query, is then provided to the LLM. By incorporating this new knowledge and leveraging its training data, the LLM is able to generate more accurate and improved responses.

In the context of Workgrid's AI Work Assistant, RAG helps summarize data points from multiple sources, such as systems, documents, and knowledge bases, into one cohesive answer or surfaces an appropriate app when an employee asks a question.

Learn more about Workgrid's conversational AI platform.