Six skills you need to become an AI prompt engineer
With the explosive rise of generative AI in the past half year, prompt engineering is the exciting new career that's growing along with it. If you want to get a gig as a professional prompt engineer, we've outlined six skills you need to hone.
With professional AI prompt engineering jobs going for $175,000 to well over $300,000 per year, prompt engineering is an attractive gig. But being a good AI prompt engineer involves more than being able to ask leading questions. You need to combine the disciplines of AI, programming, language, problem-solving, and even art to thrive on this career path.
Prompt engineering is fundamentally the creation of interactions with generative AI tools. Those interactions may be conversational, as you've undoubtedly seen (and used) with ChatGPT. But they can also be programmatic, with prompts embedded in code, the rough equivalent of modern-day API calls; except, you're not simply calling a routine in a library, you're using a routine in a library to talk to a vast large language model.
Before we talk about specific skills that will prove useful in landing that prompt engineering gig, let's talk about one characteristic you'll need to make it all work: A willingness to learn.
While AI has been with us for decades, the surge in demand for generative AI skills is new. The field is moving very quickly, with new breakthroughs, products, techniques, and approaches appearing constantly.
To keep up, you must be more than willing to learn -- you must be voracious in learning, looking for, studying, and absorbing everything you possibly can find. If you keep up with your learning, then you'll be prepared to grow in this career.
Here are six skills we recommend you hone to become an AI prompt engineer.
1. Understand AI, ML, and NLP
A key place to start is building up an understanding of how artificial intelligence, machine learning, and natural language processing actually work. If you're going to be interacting with large language models, you should understand what such a beast is, the different types of LLM out there, the types of things LLMs do well, and areas where they are weak.
This doesn't necessarily mean you need to become a computer scientist capable of creating your own LLM, but it does mean you need to understand a lot about the internals and capabilities of the tools you're trying to craft a career around. The key to this will be educating yourself by whatever means available, including traditional courseware, reading lots of articles and technical papers, attending conferences, and doing your own experiments.
One resource you should definitely check out is highlighted in ZDNET's article about a prompt engineering course. Sabrina Ortiz points you to a nine-part online class about using ChatGPT in development projects. The course is sponsored by OpenAI, the makers of ChatGPT and DeepLearning.ai, whose founder, Andrew Ng, teaches at Stanford and co-founded online learning giant Coursera. I'm taking the course and I recommend you do, too.
2. Define problem statements clearly and specify detailed queries
Fundamentally, this skill is the ability to communicate with clarity. Prompt engineering is all about how to tell the AI what you need. To do that, you need to get clear on what you want to get out of the interaction.
Here's an example. Let's assume you want to know more about Salem, the capital of Oregon. You need to be clear on at least two fronts. First, you need to explain the kinds of things you want to know, whether it's the political structure, issues of city management, traffic, or where the best donut shop is. Second, you need to be able to tell the AI that you're talking about Salem in Oregon, not the Salem in Connecticut, Virginia, or Indiana, or the witch trials in Salem in Massachusetts, or Winston-Salem in North Carolina, or any of the Salems in England, Wales, Australia, and Canada.
You'll also need to build up the skill of explaining how to set expectations for the AI, how to position it to understand the perspective it needs to use to provide value, and the context and scope of the problem you want it to solve in a given query.
Here, too, you'll need to understand the limits of various LLMs and how to work around them. For example, if you want a detailed white paper, you may need to first generate an outline, and then have the LLM write each section separately. Also, keep in mind that a clear prompt doesn't necessarily mean it's a short prompt. Longer prompts can result in more accurate and relevant responses.
The bottom line here is simple: Embrace clarity, and make sure you're able to communicate without making assumptions of understanding.
3. Be creative and develop your conversational skills
Prompt engineering is much more of a collaborative conversation than an exercise in programming. Although LLMs are certainly not sentient, they often communicate in a way that's similar to how you'd communicate with a co-worker or subordinate.
When you're defining your problem statements and queries, you will often have to think outside the box. The picture you have in your head may not translate to the internal representation of the AI. You'll need to be able to think about a variety of conversational approaches and different gambits to get the results you want.
Although I hope this isn't what you're going after, my best example of taking on conversational gambits is described in "How I tricked ChatGPT into telling me lies." My goal for that experiment was to get the AI to do something it was disinclined to do. Read through the article, and you'll see how I tried a number of creative approaches to find the conversational technique that yielded my desired results.
If you want to be a prompt engineer, experience on debate teams, negotiations, and even sales will stand you in good stead because they'll exercise those conversational, persuasion, and collaboration muscles that are so essential for eliciting desired results from generative AI systems.
4. Learn about writing and art styles, and build domain expertise
Not only will chatbots write answers for you, but they'll also often do it in the style you request. In "I used ChatGPT to rewrite my text in the style of Shakespeare, C3PO, and Harry Potter," I had more fun than any human has any right to have by asking ChatGPT to write things in the style of everything from Jane Austen to classic movie pirates. You haven't lived until you've read the preamble of the US Constitution written by a pirate!
Those examples were purely for fun and experimentation, but I also used the "write in the style of" preface for setting up my experimental Etsy store. I had ChatGPT write copy in the style of Jony Ive, whose excessively flowery descriptions of Apple products have become the stuff of legend.
For example, I used my standard Facebook icon picture and fed it to Midjourney with the prompt "cubism" and this was the result.
In addition to understanding writing and art styles, it's important for you to develop (or be able to access) the domain expertise of the area you're setting up prompts for. For example, if you're working on an AI application for auto diagnostics, it's important for you to have enough familiarity to be able to elicit the responses you need and understand if they're correct or wrong.
Oh, and here's a skill within the skill: Learn about the extensions and plugins that your favorite generative AI tools use. As time goes on, those extensions and plugins will help you do things you can't do with the off-the-shelf AI tool. So learning about and using add-ons will not only keep your skills fresh but will also let you accomplish things not otherwise possible.
5. Develop scripting and programming skills
Did you ever notice that whenever someone prefaces a phrase with "it goes without saying," there's gonna be some saying happening? In any case, it goes without saying (but I'm going to say it) that programming skills would come in handy. While there will be some prompt engineering gigs that interact merely with the chatbots, the better-paying gigs will likely involve embedding AI prompts into applications and software that then provide unique value.
While you might not necessarily be expected to write the full application code, you will provide far more value if you can write some code, test your prompts in the context of the apps you're building, run debug code, and overall be part of the interactive programming process. It will be much easier for a team to move forward if the prompt engineering occurs as an integral part of the process, rather than having to add it in and test it as a completely separate operation.
I'm a firm believer that it's much easier to be patient if you have a sense of humor. Something that's infuriating can be less toxic to your soul if you can see the essential humor in that annoyance. These generative AI tools definitely require patience. They'll completely misinterpret requests. They'll lose the thread of a conversation right when you're about to have a breakthrough. They'll completely fabricate answers that are total BS.
If you can't chuckle about some of it, you're destined to have a rough time.
That's the case with programming, as well. Every programmer needs patience. One of the biggest challenges some of my students had when starting out programming was that they couldn't accept that their code wouldn't work the first time it ran. Those who couldn't stick it out and do the work didn't complete the course. By contrast, even those who were less than inspired coders, but had the patience to try, fail, research some more, and try again were very successful.
Think of it this way. AI prompting is a mix of working with an incredibly literal computer, a willful learning model that interprets things in unpredictable ways, human team members (some of whom are even more literal than the machines), and the randomly unpredictable nature of the universe.
Patience isn't just a virtue. It's a superpower.
Some additional words of wisdom
So, there you go. I've outlined six skills you need to find success as a prompt engineer. But keep in mind that two paragraphs saying "learn about AI" isn't going to get you there. These are just rough guidelines, and it's a very individualized path ahead of you that you'll need to follow.
Embrace curiosity. The world of AI is huge and changing at work speed. Don't be content with just basic knowledge or even what you read here in ZDNET. Dive deep, ask questions, and always be curious. The more you question, the more you'll discover, and the better you'll become at getting usable results.
If I can give you one essential piece of advice, it's this: Tinker. Pick projects of your own that interest you and build something. Team up with a few friends and see what you can produce. Having some hands-on experience will take you a lot farther than a list from some guy on the internet.
Get out there and do some prompt engineering. Build some small applications. Take that course I recommended. Build some stuff. Then, not only will you no longer be someone who wants to go into prompt engineering, you'll also be someone who's done it and has something to show for it.
Do it. If you do it, you will be it.
Disclaimer: Using AI-generated images could lead to copyright violations, so people should be cautious if they're using the images for commercial purposes.