Explaining ‘RAG’ - in the way I wish somebody explained it to me

First – ‘RAG’ is acronym for Retrieval-Augmented Generation used often with Generative AI topics.

Second – it is a method to add external information to universal AI Foundation Models (basic, universal models) such as Chat GPT.

Third – it does not influence or change Foundation Model – it is not doing ‘fine tuning’ of Foundation Model’s parameters, weights etc. Rather it converts additional documentation delivered by you (pdfs, Word docs…) to the form that can be understood and used by Foundation Model answering your queries.

Fourth – ok, but how it is done internally? Information from additional documents is first converted to vectors in multi-dimensional space (called embeddings). Items positioned ‘near’ each other in this complicated space have similar meaning and this feature can be used by Foundation Model to find and use them to prepare answers.

?

So what? Why does it matter? Using this method, you can add your existing documentation to make AI ‘smarter’ – even become an expert in your desired field. The process is often automatic – many AI chats, AI applications have already feature to ‘add’ documents (… and yes … this is ‘RAG’ !). So you are adding expert knowledge without the difficult and expensive process of ‘fine-tuning’ your model. That’s a something!

?

#AI #MachineLearning #ArtificialIntelligence #RAG #TechInnovation

Below I am including some I think useful links if you would like to study it further:

-????????? https://inside-machinelearning.com/en/rag/

-????????? https://expertbeacon.com/retrieval-augmented-generation/

-????????? https://aws.amazon.com/what-is/retrieval-augmented-generation/

-????????? https://learnbybuilding.ai/tutorials/rag-from-scratch

?

Let me know if you would like to hear more about embeddings, vector spaces or just more about RAG !