David Searns | Co-CEO
OMG! I thought I was good at AI prompting… I’ve been doing it ALL WRONG!
This was the immediate reaction from one of our team members after seeing a demo of how to do AI right.
Hint: Good prompting is not easy…or intuitive.
And it’s probably not what your team is doing.
Now, if you’re happy with “good enough” AI content, you can stop reading.
But if you want AI to generate top-quality, well-researched, highly valuable content for your sales and recruiting, then read on. Because most likely, your team is doing AI wrong.
The Backstory.
This past Tuesday, our Haley Marketing team held our 2024 virtual holiday event.
All 73 of us got together online for a day filled with sharing, learning, and lots of laughing.
We reviewed our results from 2024. We discussed our vision for the new year. Each of our teams shared their proudest accomplishments. And we had wonderful guest speakers who provided some terrific education and team building.
Two of those speakers were Katie Robbert and Chris Penn from Trust Insights, and they led a workshop for our team on the RAPPEL AI prompting framework.
Personally, I’d read about the RAPPEL framework before. I had even tried to use parts of it.
But having Katie and Chris walk us step-by-step through the RAPPEL process was GAME CHANGING!
AI Can Make Amazing Cocktails?
For our workshop, we could have done something practical, like using AI to write a great blog post that engage HR managers…but what fun would that be?
Instead, our challenge was to develop new and innovative holiday craft cocktails (or mocktails for those who don’t consume adult beverages).
And I have to say, hats off to ChatGPT for coming up with an amazing Cranberry Rosemary Smash!
But this isn’t a story about bourbon drinks (although I will share the recipe later).
This is a story about what AI can really do when you use it properly.
Better Prompts. Better Results.
Yeah, you know this. I know this. Everyone knows this.
But what does it really mean?
Before yesterday, I thought that a good prompt meant the following:
- Tell the AI the role you want it to play.
- Give it a little context (background information).
- Clearly define the output you want.
- Go!
That gets you “okay” results, but not great results. For truly great results, you need a deeper prompting methodology that provides the AI with more robust instruction.
With ChatGPT 3.5, you can provide a prompt with up to 4,096 tokens – that’s around 3,500 words or 7-8 pages of written copy, depending on how you write.
That’s a lot. Probably more than you normally use in an AI prompt.
With Google Gemini, you can have an input prompt of 2,097,152 tokens.
THAT’S 3,500 PAGES OF INPUT!
Now, you’re probably not writing 3,500-page input prompts yourself. This is where frameworks come into play. You use a framework to ask questions, gather data, and then build your input prompt.
That’s what RAPPEL does.
The RAPPEL Prompting Framework
While I can attempt to explain the RAPPEL framework, you may prefer to get the details right from the sources. Check out this post to learn more:
The Trust Insights RAPPEL AI Prompt Framework.
But, in a nutshell, RAPPEL is an acronym that tells you how to prompt:
- ROLE: Tell the model who it is…and be specific!
- ACTION: What is the task you are asking the model to do? Be as precise and clear as you can about the outcome and output you want.
- PRIME: Ask the model what it knows about the topic. Priming helps you validate the model’s knowledge and gathers the background context the LLM needs to perform the task.
- PROMPT: Tell the model what to do. This is where you need very detailed, very explicit instructions about what you want done. The more detailed and precise the prompt, the better the output.
- EVALUATE: Score the model’s results and ask it to improve. This is a step 99.9% of people skip. AIs are actually capable of reviewing and even evaluating their work. You can ask the AI to critique its work and iterate toward an output with an acceptable level of quality.
- LEARN: Ask the model to encode the process. Here, you have the AI build system instructions, which is a detailed prompt, that you can then re-use again in the future.
RAPPEL is not a simple process. And it’s not entirely intuitive – some of the steps were created based on how LLMs process data. In effect, you use RAPPEL to ask questions in a way that “programs” the AI to provide the response you want.
RAPPEL (or a framework like it) enables you to use generative AI to create truly high-quality content for your sales, recruiting, and marketing.
Back to Cocktail Class
So, what did I do with RAPPEL?
- I asked the AI to be a “world-class bartender who works at a high-end speakeasy.”
- I asked it to develop a recipe for an original holiday-themed craft cocktail based on bourbon.
- I had it tell me all it knows about creating the perfect craft cocktail.
- I asked it to build a scoring system to evaluate a craft cocktail.
- I then challenged it to create my holiday drink—using no more than 5 ingredients that are all easily available at any supermarket. Oh, and I requested a drink this not overly sweet.
My ChatGPT bartender then presented me with three drink options:
- The Winter Oak
- Fireside Boulevardier
- Maple Cranberry Smash
This final one sounded like a winner to me, but it wasn’t the highest scoring.
So I asked my AI bartender to take another stab at refining that last drink to make it the top-scoring option.
I received three new variations of this (already making me thirsty!) drink, and here is the winning recipe:
Variation 2: Cranberry Rosemary Smash A fresh herbal twist adds sophistication and aromatic complexity. Ingredients
- 2 oz bourbon
- 1 oz unsweetened cranberry juice
- 0.5 oz fresh lime juice
- 0.5 oz rosemary-infused simple syrup
- Garnish: Fresh rosemary sprig and a few cranberries
Instructions
- Make the rosemary simple syrup: Simmer 0.5 cup sugar, 0.5 cup water, and 2 rosemary sprigs for 5 minutes; strain and cool.
- Combine bourbon, cranberry juice, lime juice, and rosemary simple syrup in a shaker with ice.
- Shake well and strain into a rocks glass over fresh ice.
- Garnish with a rosemary sprig and fresh cranberries.
Yes, ChatGPT just taught me how to make the perfect holiday drink. And RAPPEL taught me how to actually prompt properly.
Oh, if you’d like to see what my Cranberry Rosemary Smash looks like, here it is (image courtesy of ChatGPT and Dall-E): |
