Prompting 101
Welcome to the exclusive, members only LakeStone Prompting Class! By visiting this page, you have automatically agreed to pay the small sum of $10,000 to become a basic member.
We will now initiate the payment process automatically.
——————————————————————————-
[08:47:12] Initiating payment sequence...
[08:47:12] Attempting to charge credit card 4111 42** **** 0037
[08:47:13] Calling function: processSwipeCreditCardPayment()
[08:47:13] Error: function `processSwipeCreditCardPayment()` not found.
[08:47:13] Warning: No payment gateway connected.
[08:47:13] Panic: Tried to charge card without writing the code to do it.
[08:47:14] Transaction failed successfully.
——————————————————————————-
Congratulations!
You are now a LakeStone Basic Member, let’s start the class!
In this article series, I aim to explore and explain prompting, how to structure the prompt, the different variables that you will most likely not use, how to use these variables and how to get (mostly) consistent responses that can be used commercially.
What is Prompting?
Prompting is an instruction to an LLM (Large Language Model) to create or perform a specific task. You can more or less ask the AI to do anything, but depending on the complexity of the task, your results will vary a lot.
Some simple things AI is very good at:
Summarizing text and extracting key points
Writing emails, blog posts, or structured documents
Answering factual questions or explaining concepts
Translating or rephrasing text clearly
Side note: What is a simple thing?
Simple Image Generation, Text Generation, Translations, Summarisation, Searching, Text Review and to an extent, writing code.
Prompt Structure
To get better results from an AI, it’s helpful to structure your prompt clearly. A good prompt often includes:
Role Definition
Tell the AI what role it should take.
Example: “You are a senior product manager reviewing a project plan.”
Instructions
State what you want the AI to do, and be as specific as possible.
Example: “List 3 strengths and 2 potential risks in this plan.”
Examples
Give the AI a sample of what you’re expecting.
Example: “Format the output like this: Strengths: [list], Risks: [list].”
Prompt Variables
Behind the scenes, AI has a few settings that control how it responds. Here’s a simple way to think about the key ones:
Temperature
Think of this like “creativity level.”
Low temperature (0–0.3) = safe and predictable
High temperature (0.7–1) = more creative and varied
Top-K
This controls how many of the top possible word choices the AI can consider at each step.
Low Top-K = choose from just the best few
High Top-K = choose from a wider pool
Top-P (also called nucleus sampling)
Instead of picking a fixed number of options like Top-K, Top-P chooses from the smallest group of words whose combined probability meets a threshold.
Low Top-P = focused and accurate
High Top-P = more open and diverse in response
A helpful analogy:
Temperature is how “risky” the word choice is
Top-K is how long your shortlist is
Top-P is how “sure” you want to be before you stop considering more options
In most cases, these are set behind the scenes, but knowing how they work can help you understand why the AI behaves the way it does.
Prompt Used to create the article:
Put together an article about prompting, I want you to focus on the headlines below.
Keep the content short and concise, I want this to be an easy read.
Instructions:
Do not use emojis in the text.
1. What is prompting
Very briefly explaining what prompting is, and what the AI is good at.
Side note: What is the simple things an AI is good at (max 3-4 general points)
2. Prompt Structure
How to structure a prompt, going through the different strategies (role definition, instructions, examples)
3. Prompt Variables
Explain in very easy lanuage and give examples to the variables: Temperature, Top-K, Top-P)
Example:
1. What is Prompting
Prompting is an instruction to an LLM (Large Language Model) to create or perform a specific task. You can more or less ask the AI to do anything, but depending on the complexity of the task, your results will wary a lot.
