Prompting as a Lean Six Sigma Master

|

Said

Learning how to effectively ask ChatGPT or prompt engineering is like knowing how to effectively Google back in the early 2000s.

This is another way to show how over the years variation is manifesting in different ways and how Lean Six Sigma can be applied outside of manufacturing.

With the raise of AI and ChatGPT, the power of asking the right questions is more important than ever.

And to ask the right questions we have to think critically, a skill that I feel is getting lost with so much automation.

Here are some of the things I use to maximise the effectiveness of my prompts.

Prompting is based on Linguistics.

📖 Linguistics is the science that studies the nuances of a language and how it is used in different contexts.

3 rules of thumb I always use:

1 – I write my prompts as an SOP (Standard Operational Procedure).

I give as much context as possible, like the format of the response, the steps I want the LLM to go through to craft its response and so on.

2 – I specify a Persona for the LLM.

The models have been trained on all the data available on the Internet (ChatGPT4 looking at you mate) so I can tune my response to make the LLM weight more the data from a specific individual

3 – Providing data the LLM has never seen is like black magic.

This is a double edge sword, big disclaimer, if you use your own data you are authorizing the third party to use it to make their LLM better so beware.

But it’s extremely useful, because that way I can niche down my request in a way that otherwise it would be impossible.

Let’s put an example, let’s say I want to bake a cake and want to use the help of ChatGPT to do it.

A basic prompt would be “Can you give me the recipe for a cake?” but this prompt would be awful as there is no context.

Refer back to the Linguistics comment.

Now let’s use these 3 rules to craft the most compelling and accurate prompt.

Let’s create this SOP and give initial context:

“I want to bake a CARROT CAKE and I would like you to give me a step by STEP BY STEP LIST OF ELEMENTS TO FOLLOW in a way that MAKES THE PROCESS ENJOYABLE AND FAIL PROOF. I want you to also give me a BREAKDOWN OF THE MACROS AND CALORIES the CARROT CAKE will have.”

Let’s give it a persona to take.

“I want you to put yourself in the shoes of GORDON RAMSAY the famous chef to make this CARROT CAKE”

Check out ChatGPTs4 response below

Now the third one is a bit more complex to understand. To make it simple, I will give an extremely basic example:

“What should Said do this weekend?” → The LLM doesn’t know who Said is or what he likes doing.

But if instead I give context to the LLM…

“Said likes BASKETBALL and GOING OUT FOR FOOD”

When we ask the same question now it’s much easier for an LLM to give you an answer.

Have you tried using LLMS to make your work easier? If so, what are your best prompts? How do you make sure you get consistent responses?

Leave a Comment