RAG — Medicine for AI Hallucination

In my previous blog, I wrote about AI hallucinations, how model can sound confident yet be completely wrong. So, the obvious question is “What do we do about it?”

The answer is something that is called RAG.

What exactly is RAG? RAG (Retrieval-Augmented Generation) is the process of optimizing the output of a LLM (Large Language Model), so it references to an authoritative knowledge base outside its training data sources before generating a response.

That’s quite a long and technical definition…!!!

Let’s simplify it.

We already know that LLM generates text by just predicting the most likely next word based on patterns it has learnt from previous training data and input it receives. With RAG, something different happens, it first searches a trusted knowledge base across the web and then uses that retrieved information to generate better answer.

Here, optimizing the output of a LLM refers to making the answer more accurate, reliable and less imaginary. Instead of guessing, it responds with answer present in real time.

Authoritative knowledge base refers to trusted knowledge source like a database, documents, company files or website content.

The name itself explains everything Retrieval — Searching and fetching relevant data from a database Augmented — Improved or enhanced Generation — Responding with final answer

And that is why, if hallucination is the disease, RAG is the medicine!

Persi? Perfetto!😉