Book review: “Developing Apps with GPT-4 and ChatGPT”October 22, 2023
Whether you're pro or con the usage of AI, LLMs, and ChatGPT one cannot deny that there are a plethora of possibilities now available to us.
Now, given the rapid shift in this field, a lot of us are left with a lot of queries such as: what are the differences between the models, how do I query them and of course, how can I use them to my advantage?
That's where this useful book by Olivier Caelen and Marie-Alice Blete comes in particularly handy. It offers us some interesting insights into the various models, and how to make use of these using the ChatGPT Python library.
About the book
price: €59.99 for the eBook
publication date: September 2023
publisher: O'Reilly Media, Inc.
About the book
The book starts of with a brief introduction of the key concepts of various generative models, and the evolutions & challenges behind Large Language Models (LLMs) and Generative Pre-trained Transformer (GPT) that led us to this point.
- how predictions work
- how LLMs are trained
- the use of vector databases to retrieve complex data in a structured and efficient manner
It then takes us for a brief stroll through the history of ChatGPT, the challenges encountered, and why there are different models.
Then we consider different possible uses such as learning a new language (Duolingo), improving quality of life (Be my eyes), ...
To wrap up the introduction we take a look at the new possibilities thanks to the plugin system.
As to the meat of the book: we start of by experimenting within the OpenAI Playground, before diving into the actual fun bits.
We learn about
- prompts, the different variants & their finetuning
- the tokenization system (and the associated costs)
- how to use a variety of APIs to develop simple applications (moderation, news generation, ...)
- finetuning models
- the plugin system to add capabilities
- What (not) to do
It's a very interesting book which gives thorough, practical examples using a new & rapidly evolving tool. Pro or contra, it's a very worthwhile read, so you can better evaluate the potential for your own use-cases.
For me, it also provided some interesting insights into how to avoid/reduce hallucinations/finetuning my queries, feed it more (up to date) data, and how to provide extra functionality to create a better user experience. It's certainly given me a lot of food for thought for my own "Simon says" application.
I was lucky enough to meet Marie-Alice at Devoxx where she gave a very interesting talk, so if you want to get some more insights I highly recommend watching it:
In case Java is more your forte than Python, there's also the useful langchain4j library.