Making killer prompts
Prompts are the text instructions that users provide to generative AI systems to tell them what kind of output to generate. For example, "Write a poem about the British countryside" or "Generate an image of a cyberpunk city."
In simple language it’s literally what we type into ChatGPT.
Carefully engineering prompts is key to guiding the AI to create high-quality results.
Prompts start as text but can help us generate imagery, audio and video.
Here for example I used the prompt “Disney Snow White, [country name], photorealistic” to create a thread about Snow Whites from all over the world:
What is prompt engineering?
Prompt engineering is the process of iteratively refining and optimizing prompts to get the best possible output from an AI system. It involves strategically structuring prompts, providing relevant context, and giving clear instructions to guide the AI. Prompt engineering is key to utilizing generative AI effectively.
If you use the prompt:
Write a blog article about yoga
then the result will be crap.
ChatGPT will create a generic version of that blog post.
It will read like an AI wrote it.
However if we adjust our prompt to something like this:
Act as an SEO expert blog writer with extensive yoga experience
Outline 5 tips, moving from beginner to advanced for improving yoga through self study. Create a yoga themed acronymn for the 5 part framework.
Use the AIDA copywriting framework.
Each tip should include between 3-5 steps.
End with a call to action to sign up to the newsletter.
Provide references and additional resources
Create internal links between articles.
Each article should be 1000-1500 words long.
The resulting output will be far superior.
Why Prompt Engineering is Needed
AI is meant to be smart right? Why do we need prompt engineering at all?
AI systems like ChatGPT do not inherently understand human intent. Their capabilities are limited by their training data.
Like with the ChadGPT vs. ChatGPT example above we have to actually tell it what we want and give specific context information.
By engineering effective prompts, we can get better results from AI and overcome some of its limitations.
Some key reasons prompt engineering is needed:
AI has blindspots outside its training data
Models can “hallucinate” (make stuff up) incorrect facts
Outputs may be generic without clear instructions
Providing context and examples guides the AI
Basically well-designed prompts act like guardrails, steering the AI toward high-quality responses. As we get better at prompting we’ll get better results.
Overview of Prompting Techniques
There’s a huge range of prompting techniques. Here are some of the most useful:
Instruction Prompts: Giving clear step-by-step instructions to guide the AI
Role Prompting: Asking the AI to respond as a specific persona or character
Few Shot Prompting: Providing examples to illustrate the desired output
Priming: Setting an initial context to shape the conversation
Embedding Prompts: Importing knowledge and logic from an external source
Chaining: Breaking down complex prompts into a series of smaller prompts
Constraining: Adding constraints to reduce undesired responses
Human Feedback: Iteratively collecting human judgments to refine the prompt
Checklisting: Providing a checklist to verify quality and completeness
Simplifying: Reducing prompt complexity to isolate issues
Dialogue: Conversing with the AI to guide it interactively
Phew that’s a lot. And there are more.
Thankfully you don’t need to worry about this too much.
Prompt engineering is the skill I’ve used to build the prompts and workflows you’ll be working with as you work through the course. I’ve already done the hard work for you using these techniques and more.
That said, as you work through the course you’ll naturally start to pick up what makes a good prompt. You’ll be following along, using and editing my prompts to your needs. Just doing this will give you an intuitive idea of what good prompt look like and how powerful they can be.
You’ll be learning by doing.