1 Big Mistake I Made When I Was First Learning about AI Prompting
Learn from my AI Prompting mistake
Like most beginners, I made a lot of mistakes when I first started learning about AI.
But this was the biggest one, by far: One Sentence Prompts
Here’s what happened:
My first interaction with AI was asking an AI chatbot single sentence questions so I thought using single sentences is how we interact with AI.
Some might consider my one sentence prompts “zero-shot prompting”, where you interact with a model without providing it additional context or examples, and little did I know then that was what I was doing. I didn’t know there was a methodology to prompting. But, it’s also worth acknowledging that making this mistake taught me a ton.
I learned that there is a whole discipline to prompting to get the most out of AI.
Try this Prompt Structure:
Prompt = Persona/Role + Context + Task + Example + Output Format + Tone
- Persona/Role: Instruct the model to take on a character or identity, such as grade school teacher, New York Times editor, or financial analyst which will give the LLM a blueprint for the expertise you are looking for.
- Context: Provides extra details and background information so the LLM has all the information it needs to do its best work.
- Task: The thing you want the LLM to do.
- Example: Examples teach the model what a good output looks like.
- Output Format: Specify if the output should be formatted a certain way such as JSON or XML.
- Tone: Specify if the output should be formal, casual, or somewhere in between.
Lesson Learned:
Only Task is mandatory but the more elements of this structure you use, the better the AI’s response should be.
After learning this, I realized that I was only providing the Task in my one sentence prompts. According to Google’s Workspace Prompting Guide, the most fruitful prompts average around 21 words with relevant context yet my prompts were probably less than nine words. Prompting is evolving so this structure could change but now I know I need to provide more than just one sentence to the LLM and I should have some sort of structure.
This is why I encourage everyone to see their mistakes (and “failures”) as necessary steps along the path.
There is always a lesson to be learned and prompting is a skill we can all learn.
References
- Gemini for Google Workspace Prompting Guide 101 — October 2024 Edition
- Google Overview of Prompting Strategies
- ChatGPT Prompt Engineering for Developers
Notes
- Special thanks to my coworker Rihana Msadek who shared this prompt structure with me.
