Right here is an LangChain normal immediate for RAG:
You might be an assistant for question-answering duties. Use the next items
of retrieved context to reply the query. If you do not know the reply,
simply say that you do not know. Use three sentences most and hold the reply
concise.
Query: {query}
Context: {context}
Reply:
Do you write your prompts in an analogous approach? Hopefully not.
This isn’t even a place to begin. That is nothing.
From a immediate engineering standpoint, this immediate lacks just about every little thing that makes a superb immediate: Few examples, output format, edge case dealing with, reasoning, limiting responses to what’s in context, and far, rather more.
Lots of people nonetheless assume that the immediate doesn’t have to be that good, I simply want good information, a superb RAG setup, a superb framework, good code, then it’ll work. Nope! In fairly complicated circumstances, and past POCs, you may get 80%, 90%, 95% accuracy per information level with this method. With some luck. What you’re going to want for many techniques operating in a enterprise context is 98%, 99%, 99.5% accuracy. And also you’ll by no means get there with prompts just like the one above.
Here’s a comparability of an actual immediate template with the instance template above (I needed to blur the true immediate):
