You're reading from Unlocking Data with Generative AI and RAG Enhance generative AI systems by integrating internal data with large language models using RAG

Product type Paperback

Published in Sep 2024

Publisher Packt

ISBN-13 9781835887905

Length 346 pages

Edition 1st Edition

Concepts

GPT/LLMs

Author (1):

Keith Bourne

View More author details

Table of Contents (20) Chapters

Preface

1. Part 1 – Introduction to Retrieval-Augmented Generation (RAG) FREE CHAPTER

2. Chapter 1: What Is Retrieval-Augmented Generation (RAG)

3. Chapter 2: Code Lab – An Entire RAG Pipeline

4. Chapter 3: Practical Applications of RAG

5. Chapter 4: Components of a RAG System

6. Chapter 5: Managing Security in RAG Applications

7. Part 2 – Components of RAG

8. Chapter 6: Interfacing with RAG and Gradio

9. Chapter 7: The Key Role Vectors and Vector Stores Play in RAG

10. Chapter 8: Similarity Searching with Vectors

11. Chapter 9: Evaluating RAG Quantitatively and with Visualizations

12. Chapter 10: Key RAG Components in LangChain

13. Chapter 11: Using LangChain to Get More from RAG

14. Part 3 – Implementing Advanced RAG

15. Chapter 12: Combining RAG with the Power of AI Agents and LangGraph

16. Chapter 13: Using Prompt Engineering to Improve RAG Efforts

17. Chapter 14: Advanced RAG-Related Techniques for Improving Results

18. Index

Why subscribe?

19. Other Books You May Enjoy

Take your shot

No-shot, single-shot, few-shot, and multi-shot are common terms you will hear when talking about your prompting strategy. They all stem from the same concept, where a shot is one example you give to your LLM to help it determine how to respond to your query. If that is not clear, then I could give you an example of what I am talking about. Oh wait, that is exactly the idea behind the shot concept! You can give no examples (no-shot), one example (single-shot), or more than one example (few-shot or multi-shot). Each shot is an example; each example is a shot. Here is an example of what you would say to an LLM (we could call this single-shot, since I am only providing one example):

"Give me a joke that uses an animal and some action that animal takes that is funny.
Use this example to guide the joke you provide:
Joke-question: Why did the chicken cross the road?
Joke-answer: To get to the other side."

The assumption here is that by providing that example, you are helping guide the LLM in how you respond.

In a RAG application, you will often provide examples in your context. That is not always the case, as sometimes context is just additional (but important) data. However, if you are providing actual examples of questions and answers in the context with the intention of directing the LLM to answer the new user query in a similar manner, then you are using a shot approach. You will find that some RAG applications follow the multi-shot pattern much more closely, but it really depends on the goal of your application and the data you have available.

Examples and shots are not the only concepts that are important to understand in prompts, as you will also want to understand the difference in the terms referring to your approach to prompts. We will talk about these approaches next.