What is RAG in AI?

RAG- Retrieval Augmented Generative- is a relatively recent addition to Generative AI architecture, the term being coined by Patrick Lewis in his 2020paper¹. But what is it, and how does it work? MISSION inc. with Monty AI explain all! 

Early chatbots such as Gen 1 Xiaoice² or the infamous TAY³ were trained on a large volume of data, which could be updated by users’ query data. Their ability to deal with a query outside of their source dataset- no matter how large- was, however, limited. They could either find and return an adequate answer, or they failed entirely. This is still seen in simple chatbots giving a robotic and frustrating stock response such as “I’m sorry, I don’t have access to that information. Is there anything else I can help you with?”

RAG allows an AI model to query a dynamic data source (or multiple data sources) such as vector databases or RSS feeds. The benefit of this is clear; you can expand the data which the AI will use as a source of truth. This also introduces a temporal element; stale or incorrect data can be updated / replaced with the more relevant and accurate information.

Another less obvious but also highly valuable use case is where a business wants to interrogate proprietary, confidential data and glean insight, but also wants that data protected and ringfenced. Your data can be kept securely in a vector database such as Weaviate or Pinecone, and you retain ownership of that data, ensuring only authorised access is granted.

MISSION inc. have undertaken this approach with Monty AI, the in-house chatbot built to query your company and business data, in a method that utilises retrieval of data from a vector database before running the user query through a LLM. Using generative AI to interpret the data and give business leaders insight into the company performance allows large quantities of data to be parsed, saving time and finding novel solutions to problems that might not even be apparent at the human level.

 To learn more about how MISSION inc. are transforming business data with their intuitive KPI dashboard and Monty AI, sign up today for a free trial.

References

¹Lewis P, et al. 2020. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. Proceedings of the 34thInternational Conference on Neural Information Processing Systems. Curran Associates Inc., Red Hook, NY, USA, Article 793, 9459–9474.

² Zhou L, et al. 2020. The Design and Implementation of XiaoIce. Computational Linguistics 2020; 46 (1)

³ Summers N, 2016. Microsoft's Tay is an AI chatbot with 'zero chill'. Online, URL: https://www.engadget.com/2016-03-23-microsofts-tay-ai-chat-bot.htmldate accessed 20240612