6 Comments
User's avatar
Jenny Ouyang's avatar

You’re taking a really different approach from how I built my RAG, and I love it!

The way you’re handling hallucination is super considerate. Definitely something I want to try in my own setup.

Claudia Ng's avatar

Thank you, Jenny!

Chris's avatar

Do you think OpenAI will bake this into version 5, it seems like a no brainer?

Khe Hy's avatar

Claudia, the timing couldn't be more perfect! I just texted a friend:

Btw — if you come across some “understanding RAG” without being technical content/projects , pls pass along

Claudia Ng's avatar

Haha nice! Hope this clarified things for you. Feel free to ask if you have follow-up questions after reading this

Claudia Ng's avatar

RAG (Retrieval-Augmented Generation) isn’t something that gets “added” to an LLM, it’s an architectural pattern. It retrieves relevant data (from a vector store, database, etc.) and feeds it into the LLM as context. The model then generates responses using that extra information, but the retrieval system lives outside the model itself.