Book review: Developing Apps with GPT4 and ChatGPT

I received a free copy of this book thanks to their authors Marie-Alice Blete and Olivier Caelen and I finally read it. It is a very good book that accomplishes the task of going in depth with the fundamentals of GPT4 and ChatGPT, explaining the key concepts and how they work under the hood.

What really stood out? As a Site Reliability Engineer with a focus on Observability, I really liked the ethical perspective of this book when describing what LLM are capable of, how to use them, and what future they hold ahead. Pointing this out right from the box, it's not just another coding book. It aims to educate, to look at the big picture of apps like ChatGPT, artificial intelligence and what they are in a nutshell: tools in our hands.

Let's deep dive a bit into this not so thick book, but lengthy enough to give us proper instructions on how to operate GPT4 and ChatGPT.

The book rigorously starts with GPT4 and ChatGPT fundamentals, what artificial intelligence, machine learning, deep learning and transformers have in common, what NLP is and what it its objective. I really liked the part when the architecture of transformers was clearly explained with examples, the history of GPT, what are tokens used for and what are the limitations of AI. Probably, if you are an avid reader on the subject, you already are familiar with many of these concepts, but the book makes a good job and explaining this more academically and visually satisfying.

Chapter 2 goes further into the AI landscape and dives into GPT4 and ChatGPT's APIs. If you are a bit of an academic nerd like myself, you will appreciate the clarity and how organically the book deals with information. It explains the available models for OpenAI, how to use Playground, OpenAI's Python library and goes into showing us how to use GPT4 and ChatGPT. Nice! I really liked the part with the field description and their examples. It didn't leave time for confusion. The highlight of this chapter were its considerations: tokens and price, security and privacy. Right on the spot!

Chapter 3 explores the app development part: how to use the API keys, a few software architecture design principles, and the inevitability of prompt injection. This chapter is rich with project examples, be sure to give it some time.

Chapter 4 is by far my favourite, mainly because I liked the role of effective prompts and how big of a role the human plays in this part. GPT4 and ChatGPT are not plug and play. They need our continuous guidance. Just like children, we guide them through a series of steps and instructions, perfecting its learning along the way. We do need to know how to use it responsibly and these apps will be as good as we can be as agents of effective prompts. There is a great distinction as well between fine-tuning and few-shots learning, their pros and cons and applications.

Chapter 5 is focused on the LangChain framework, its modules and plugins. The book kind of ends abruptly, probably because I was expecting more to come. Not a fault of this book. I definitely wish to see a sequel but all good things come to an end and I am happy I have learnt so much from this book.

To wrap it up: Developing apps with GPT4 and ChatGPT is definitely not to be missed if you are taking off on this subject or going a bit deeper. There is so much to cover on the subject and be sure to explore your curiosity with real life attempts to play with ChatGPT, OpenAI and actually build your own app with these tools.