What is Prompt Engineering?
Prompt engineering is a new term quickly working its way into the world of AI. You may be wondering what this new title that people have started giving themselves is and why there is so much discussion about it. So let’s get into what it means, how it's done, and why you should even care!
Prompt engineering is the practice of creating prompts for AI models that give the user their desired output. In other words, a prompt engineer looks for the most effective input they can give an AI model to achieve a high-quality, useful response.
An important part of being a prompt engineer is understanding where an AI model might fall short and what its capabilities, especially the less obvious ones are. This can help users extract what they need from an AI model more effectively. Understanding the hidden strengths of an AI model can also allow users to leverage AI models in ways that their competitors may not be doing.
How do you design effective prompts?
There are a variety of factors that go into creating high-quality prompts but as a general rule, you should always be specific, concise, and provide context. The more specific your request, the less likely the program will be to mistake your intentions and generate a response that doesn’t answer your question.
Even if you are trying to accomplish a complex task, being concise is important as well. The best way to do this is to break up your larger task into sub-prompts. Instead of trying to accomplish multiple tasks at once, this will make the program create a concrete response for each sub-prompt. This also allows you to course correct if the program doesn't generate your initial intended response or you can have the program dig deeper into certain responses it gives you. You can then tailor your subsequent prompts based on feedback from the program.
Finally, giving context can be the most impactful way to get highly effective responses. One way to give context is to define the program’s role. Imagine you want responses in the tone of a certain character or from the perspective of a specific type of professional. You can tell the program to take on a certain role to help it frame the kind of responses you are looking for.
Another way to give context is to describe your target audience for the content being generated. If you are creating marketing content aimed at young professionals in a specific city, within a specific age range, or any other demographic, you should specify that in your prompt.
Finally, you should tell the program the channel that you are delivering content through. If you want a long-form blog post vs an e-commerce product description, specifying that will ensure that the output you receive fits any constraints you might face and will signal to the program how it should format its response.
As the old adage says, you get out what you put in.
Now that we know how to create effective prompts, let’s look at some examples of prompt engineering use cases.
Let’s say you want a chatbot to create content while mimicking either your own style of writing or someone else’s. You could create a prompt that includes sample text of your own or someone else's and then tell the chatbot to start by analyzing the provided text. The next prompt could then be to write about a certain niche topic in blog post format using that writing style. This creates a context for the chatbot to work within, both in terms of writing tone, as well as the content type you want to be provided. It also breaks the overall request into smaller, more concise steps.
You could have an AI program act as a recruiter. You could literally tell the program, “I want you to act as an internal recruiter for my coffee company. Come up with a job description for the marketing manager role that includes, tasks x, y, and z. Also, include requirements a, b, and c in the job listing.” Once the program generates a response, you can then follow up and say “Next, create a list of five questions that you would ask a candidate for this job”. This would allow you to get a job posting that you can tweak for specifics as needed and on top of that be equipped with questions for completing the interview itself.
You could even flip this prompt and pretend you are the job candidate! You could tell the program to take on the persona of a hiring manager at a coffee company who is hiring for a marketing manager role. You could then have it ask you questions one by one and have the program provide you feedback on your answers as you go to sharpen your interviewing skill for this specific role.
Another great prompt would be for lawyers or anyone who needs to look at a debate from multiple angles. You could prompt a chatbot by saying, “Here is a trial and these are the facts of the trial. I am arguing on behalf of the defendant. You will take the role of the plaintiff’s lawyer. Here are my top three arguments for this case. Based on the facts provided, provide a counter-argument to each of my points.” You could then figure out what your counter-argument to the chatbot’s counter-argument would be and keep iterating on your defense.
How can I become a prompt engineer?
This question might be on a lot of people’s minds and will certainly be on many more people’s minds going forward. The good news is that if you are asking yourself this question now, you are early in the game and can get a jump on your prompt engineer journey well before the competition.
The best steps you can take now to become a prompt engineer are to experiment with different AI programs by interacting with different companies’ chatbots, image generators, etc. Practice entering prompts and figure out which ones give you detailed and useful responses and which cause the program to fall a bit flat in its output. Be sure to also experiment with different types of prompts. This means that you shouldn't always just ask a chatbot to generate one type of content like composing an email. Instead, experiment with what you can get it to do. Ask yourself, “What tasks from my daily life can I delegate to an AI program?” and then practice doing that.
Another way to learn about prompt engineering is through online courses. There are people and companies offering courses and certifications for prompt engineering. While some of these may be useful and valuable, be aware that this is a relatively new discipline and the people trying to sell you a course may not be much farther along in their learning than you are. There is a ton of free content around prompt engineering so start by following some interesting profiles on Twitter or reading message boards for inspiration on prompts to try. Here is a thread to get you started.
AI programs, especially chatbots, can save you literal hours and can give you high-quality responses when you interact with them correctly, as we’ve shown in this post. The majority of people still have not used these products and likely won't for many months or even years to come. You have a unique opportunity to not only give yourself an edge by implementing these tools and practices in your life and business, but you can also become a subject matter expert. Once everyone else gets on the AI train you’ll be miles ahead and will have increased the value of your skillset making you more competitive as a job candidate, business owner, or maybe you’ll have just learned how to make your own life easier. No matter how you look at it, you win!
— AI Review Guy