What is Retrieval Augmented Generation?
Mattias Acosta
Revolutionizing Workflows with AI Systems | IBM Solutions Engineer | Systems Architect
Retrieval Augmented Generation — this AI technique uses an external authoritative data source in order to enhance a an LLM's response with accuracy and relevance to the particular task at hand.
RAG will extend the powers of an LLM by giving it specific data to generate a response. Using the RAG technique will ensure that the LLM is informed by the knowledge source of your choice.
How it functions sequentially:
The owner of the data converts the specific data source you want to use into a vector database.
The user creates a prompt.
The RAG takes the user prompt and uses it to search the specific data source you have provided it with.
The RAG performs a relevancy search - the prompt is converted to a vector representation and mapped to the source’s vector database.
Then the prompt is augmented - RAG will use the data it found to create a new augmented version of the prompt.
The augmented prompt is then fed into the LLM to generate a more accurate and contextualized answer…
Properties of RAG:
Citations - RAG provides reference sources that the user can look up.
Relevancy- the RAG data source can be: live social media feeds, news sites, or other frequently updated info sources.
领英推荐
Privacy - Because RAG only uses the data sources that you select, it ensures that sensitive information within the org is not accessed and used to generate responses.
Efficiency - since the initial retrieval phase narrows down the context and reduces the amount of data that needs to be searched, RAG models are more efficient.
Use Cases:
Company Knowledge Base - The most common use case here is for a company to use a knowledge base in order to generate answers for customer service. This is an area for a lot of growth: as company’s leverage AI with customer service, they can use RAG to give agents very specific context.
RAG-Powered Textbooks - Another cool use case is studying with this. I would love a textbook that has an LLM with RAG connected to it and allows me to ask it questions. Of course, there are risks here, and I would have to go back and look through the text to ensure 100% accuracy of any answers generated… but many RAG’s today provide the source (or page number) that it retrieved its answers from.
Bio
Mattias is a systems thinker who drives innovation and builds character. He is also well-versed in Cybersecurity and can help you protect your organization's AI models. Contact him today to chat!
Sources:
Aerospace Engineering Student at University of Illinois at Urbana-Champaign
9 个月Informative article!