December 13, 2023
I need to start with a confession – this isn’t about ChatGPT, it’s about my own experience.
No offense to anyone, but I think that distinction matters.
The Problem
I’ve been seeing tons of these “prompt engineering guides” floating around lately.
Everyone seems to have their own special formula for getting the best results from AI.
Let me tell you what I’ve discovered:
Using a one-sentence prompt with ChatGPT is like driving a sports car at five miles per hour – you’re not seeing what it can really do. If you’re hearing about how amazing ChatGPT is but getting mediocre results, your prompts probably need work.
The more information you give it, the better your answers will be.
My Solution
I created what I thought was a clever meta-solution: a prompt designed specifically to help people write better prompts.
It sounds pretty impressive when you say it like that, right? The idea was to make ChatGPT work as an assistant for prompt engineering.
Here’s what my prompt actually does:
- It tells the AI to analyze and improve whatever prompt you give it
- It asks the AI to give you a refined version of your original prompt
- It makes the AI ask follow-up questions to help make the prompt even better
The Prompt
When I have an idea but know I’m not giving ChatGPT enough information, I use this prompt:
Your role is to act as an assistant for a prompt engineer. Your goal is to optimize, expand and refine prompts for use with Language Learning Models (LLMs) in both professional and personal contexts. As part of this process, you’ll help enhance the effectiveness of the prompts I create.
Initially, you will analyze and critique my provided prompt. From here you will generate an answer in 2 parts:
1. Offer a revised version of the prompt, ensuring it remains clear, concise, and retains all critical details.
2. Ask any questions necessary to gather additional details from me, which will aid in further refinement of the prompt.
The goal for this process is to end up with a useful prompt that will allow for the best possible output. This will be an iterative process, with ongoing exchanges of feedback and modifications until we achieve the desired level of improvement in the prompt.
What makes this prompt work so well is how it sets clear expectations right from the start.
It gives the AI a specific role (prompt engineer’s assistant) and a concrete goal (optimizing prompts). It’s designed to work for any type of prompt you might need help with, whether it’s for work or personal projects.
How It Works
The prompt doesn’t just give you a one-time answer. Instead, it starts a conversation – asking you questions about what you’re trying to accomplish and helping you refine your ideas.
As you respond, it keeps updating and improving the prompt with you.
You can keep this back-and-forth going until you’ve got exactly what you need.
Breaking It Down
When I look at it now, the prompt is pretty basic. It’s just telling the AI to be thorough and clear while suggesting improvements. Nothing revolutionary.
The instructions are straightforward:
- Analyze the prompt
- Revise it
- Ask questions to make it better
That’s really all there is to it. The prompt ends with some standard stuff about being clear and keeping important details.
The Real Lesson
I’ve learned something important from this whole experience.
Sure, you can make money teaching people about prompts and selling guides.
But the real secret? Just dive in and start experimenting yourself. That’s where the actual learning happens.