“I’m using it wrong?”

Potentially. Rubbish in = rubbish out 🗑️

Let’s start with a simple thought experiment, if you wanted a colleague or an employee to do something well, you’d provide them with a clear set of instructions, wouldn’t you? The concept is exactly the same in this situation.

Utilising Generative AI Chatbots optimally is one of the easiest ways to enhance your workflow using the latest AI technology, whether you’re using free tools like ChatGPT 3.5/Google Gemini or paid plans like ChatGPT Plus and Microsoft Copilot Pro.

Today’s newsletter focuses on the art of Prompt Engineering. By the end, you’ll understand:

• The different types of prompting.

• How to craft any prompt for any Large Language Model (LLM) - ChatGPT is one of these.

• What an ideal prompt looks like.

“What are the different types of prompts?”

I’m glad you asked!

The framework I’ll explain involves N-Shot Prompts, where N is the number of examples of the ‘ideal response’ that you provide to the LLM.

• Zero-Shot Prompting:

  • No example provided. N=0.

  • Best for brainstorming, asking a question, or when you don’t have any specific examples for your desired response to a task.

  • “Determine the sentiment of this sentence: I love sunny days and ice cream.“

• One-Shot Prompting:

  • One example provided. N=1.

  • Most efficient when you have a single example that closely represents the task or output you desire.

  • “Translate these sentences into Spanish. Example: English: ‘The apple is red.’ Spanish: ‘La manzana es roja.’ Now translate: ‘The sky is blue.’”

• Few-Shot Prompting:

  • More than one example. N>=2.

  • Ideal for longer or more complex tasks.

  • “Classify these animals into ‘mammal’ or ‘bird’: Example 1: Animal: ‘Elephant’ Type: ‘Mammal’ Example 2: Animal: ‘Eagle’ Type: ‘Bird’ Now classify: ‘Dolphin’”

Understanding the different benefits and uses of these prompt types will get you halfway there, but the ‘secret ingredient’ is implementing a repeatable prompt structure to clarify your expectations and improve the LLMs understanding.

“So, what does this structure look like?”