Crafting thought: A primer for creative prompting with LLMs
7 foundational principles for crafting the perfect prompt
Thought: A New Creative Medium
The advent of LLMs has opened up a new creative frontier that avails vast computational and data resources to human thought and intention. It presents us with a new medium that weaves knowledge, desire, intent, and reason into a novel result; something akin to designing a thought process.
This is ‘prompt-craft,’ and it is a new skill that we will all learn over the coming years. This blog post is a primer to introduce some of the general concepts in this new paradigm, and then we’ll put these principles into action to craft an effective prompt.
What is “prompt-craft?”
Whereas prompt engineering is concerned with the specific techniques and mechanics of prompting LLMs, prompt-craft can be understood as the creative implementation of these techniques in pursuit of goals that are informed by the new scope of what is possible. The architect to the engineer.
Understanding LLMs: Your New Creative Partner
At their core, LLMs like GPT-4, Mistral, or Bard are AI models trained on vast amounts of text data. These models can generate text, answer queries, and even engage in creative tasks, all based on the prompts they receive. Think of them as incredibly versatile digital assistants, always ready and eager to take on any task.
However, these large models don’t actually “know” anything. Instead, they are more like strategies for transforming information as they go, based on whatever data they were trained on. This means that sometimes these LLMs will make errors in reasoning, stray from their task, or simply make things up.
Therefore, learning how to work with LLMs is not just about understanding your problem, but also about designing a strategy for problem solving that will produce a useful result.
The Art of Prompt-Craft: Designing Thought
Prompt-craft is the skill of conceptualizing and crafting effective prompts that guide these LLMs. Think of it like sculpting the thought process of the AI, making prompt-craft a crucial skill for anyone engaging with LLMs. It’s a skill that requires and rewards creativity.
Let’s explore some general concepts:
1. Be Precise
Success with LLMs starts with being clear about what you want. Whether you’re looking for innovative ideas, solutions to a problem, or creative inspiration, the way you word your prompt is everything. Every single word will factor into the LLM’s output.
2. Context is Your Canvas
Just like in any good story, providing context is crucial. Context gives the LLM useful information so that it can produce more relevant results. For context, think in terms of the Who, What, Where, When and Why that would contextualize your prompt. However, too much detail can lead to confusion, so the key here is providing or retrieving only the necessary contextual information required for the task at hand.
3. Embrace an Iterative Approach
Think of your first prompt as the starting point. If the response isn’t quite what you were hoping for, be prepared to tweak your prompt, add details, or approach the question from a different angle. Re-designing your original prompt is often better than trying to build on a sub-par response from the LLM.
4. Get Involved in the Process
Prompting at its most powerful is a collaboration between you and AI. Interactive dialogue is an excellent way to dig into a topic or question and reveal new ideas, or interrogate a proposed solution.
Incorporating role-playing in your prompt is an extremely effective technique to fine-tune this interaction. And don’t overlook the power of providing the LLM with relevant information about yourself to help it better cater to your needs.
5. Guide AI With Templates and Examples
Templates and examples serve as frameworks that guide the AI’s responses. They can be particularly handy when you need structured outputs like a list of ideas, a story outline, a set of instructions, and so on.
Structuring the LLM’s output with templates or examples can also allow you to create a sequential pipeline where one prompt’s output is perfectly tailored to serve as another prompt’s input.
6. Embrace Experimentation
The field of LLMs is dynamic and continually evolving. Experiment with different styles and structures. This exploratory approach will help you familiarize yourself with these new tools in a more intuitive way. Experimentation may also yield unexpected results, and will allow you to carve out your own signature approach to prompt-craft.
7. Re-consider the Scope of the Possible
Incorporating AI into your workflow shouldn’t replace your unique voice; it should enrich and empower it. These tools open up new possibilities and perspectives, offering a fresh lens through which to view your work and your goals. They should be approached as a means of extending your vision and your reach.
Prompt-craft is as much about expanding the larger goals that you can imagine as it is about mastering the technical prompt engineering that you use to reach them.
Prompt-Craft In Action
Let’s try creating a prompt together. Imagine you want to brainstorm ideas for an innovative product. Your prompt could be:
- Prompt:
I’m looking for unique ideas for an eco-friendly product that can help reduce household waste. Please suggest some innovative concepts, focusing on sustainability and user convenience.
This prompt is clear, provides context, and guides the AI towards the type of ideas you’re looking for. But its results will be generic, and may not be particularly useful to you personally.
Enhancing our prompt with context and interaction
Now, if we wanted to improve this prompt, we could apply several of the general concepts we’ve already explored. Let’s give the prompt a bit more context, and use interactive role-playing to let us refine our results through dialogue with the LLM.
- Prompt:
I am a young solo entrepreneur with a background in chemistry. I’m looking for unique ideas for an eco-friendly product that can help reduce household waste. You are an expert startup advisor specializing in bringing new products to market. Ask questions to help me ideate and develop innovative concepts, focusing on sustainability and user convenience.
This second approach gives the LLM more useful information about our expertise and intent, and invites the LLM to engage with us to discover a result that might be more relevant to us.
Don’t let your prompts stay prompts
Let’s give the LLM some template information that can guide our conversation toward producing a useful final product.
- Prompt:
I am a young entrepreneur with a background in chemistry. I’m looking for unique ideas for an eco-friendly product that can help reduce household waste. You are an expert startup advisor specializing in bringing new products to market. Ask me questions to help me ideate and develop innovative concepts, focusing on sustainability and user convenience. Use my answers and your expertise to produce a go-to-market strategy document for our new product.
Invoking the general template of a go-to-market strategy will inform the sorts of questions that the LLM will ask us, and will lead to the creation of a document that we can then use to bring our brainstorm into reality.
This is a well-crafted prompt for the task at hand.
Of course, this is just one simple example of the principles of prompt-craft in action. The path to mastering prompt-craft will require a deeper dive into the various prompt engineering techniques and tools available, as well as their individual strengths and weaknesses.
Your Interface to the Future
The art of prompt-craft with LLMs is more than just a technical skill; it’s a new design space. By learning to design the AI’s thought process through effective prompts, you’re not only directing the LLM but also exploring new dimensions of human creativity.
So embrace this new tool, experiment with your prompts, and expand the horizons of your imagination to encompass the new possibilities afforded to us by AI and LLMs.
And make no mistake, prompt-craft is not a trivial skill. The AI-powered future is fundamentally interactive; we will decide the future, and we will do so through prompt-craft.
— dxxmsdxy