RAG 2.0 (contextual.ai)

I can’t wait for RAG 3.0 on the blockchain.

Is this fine-tuning with a more hyped naming? Let's say a company produces 100 documents per day that are appended to a knowledge base. What's the cost and delay to be able to ask questions about these docs whenever a single one is added in?

Fine-tuning is not RAG. It is the opposite of RAG. Fine-tuning is what one is trying to avoid when using RAG. That's not to say that fine-tuning is bad; it's not, but it's just not RAG.

The article explains very little about how it works, other than “end to end.” Basically it just claims higher benchmarks. Is there a paper?

It sounds like a downside will be that you can’t mix and match. You’ll have to use their LLM and their way of creating the embeddings.