Developing Apps with GPT-4 and ChatGPT

Chapter 4. Advanced GPT-4 and ChatGPT Techniques

Now that you are familiar with the basics of LLMs and the OpenAI API, it’s time to take your skills to the next level. This chapter covers powerful strategies that will enable you to harness the true potential of ChatGPT and GPT-4. From prompt engineering, zero-shot learning, and few-shot learning to fine-tuning models for specific tasks, this chapter will give you all the knowledge you need to create any application you can imagine.

Prompt Engineering

Before we dive into prompt engineering, let’s briefly review the chat model’s completion function, as this section will use it extensively. To make the code more compact, we define the function as follows:

def chat_completion(prompt, model="gpt-4", temperature=0):
    res = openai.ChatCompletion.create(
        model=model,
        messages=[{"role": "user", "content": prompt}],
        temperature=temperature,
    )
    print(res["choices"][0]["message"]["content"])

This function receives a prompt and displays the completion result in the terminal. The model and the temperature are two optional features set by default, respectively, to GPT-4 and 0.

To demonstrate prompt engineering, we will return to the example text “As Descartes said, I think therefore”. If this input is passed to GPT-4, it is natural for the model to complete the sentence by iteratively adding the most likely tokens:

chat_completion("As Descartes said, I think therefore")

As a result, we get the following output message:

 I am. This famous philosophical ...