BackTalking to Chatbots Is Now a $200K Job. So I Applied.
Welcome to the world of prompt engineering, where you’re paid to get the best answers from AI.
PremiumMy father was a prompt engineer, like his father before him. I come from a long line of people who toiled day and night, chatting with generative-AI chatbots.
OK, not true. Prompt engineering is a totally new job that would have sounded crazy even a year ago. But it can pay six-figure salaries to people who extract the best results from the mysterious artificial-intelligence black boxes that are now part of daily life.
A year ago this week, OpenAI released ChatGPT. Not since the iPhone had I been this wowed by new technology.
Me: How do I keep squirrels out of my bird house?
ChatGPT: Install a baffle on the bird house pole, which acts as a physical barrier to prevent squirrels from climbing up. Place the bird house away from trees, fences, or buildings to minimize jumping-off points for squirrels. Additionally, consider using a birdhouse with a small entrance hole that’s too tiny for squirrels to fit through.
You had to see the bots’ ability to spit out humanlike answers to believe it. And even then, maybe you shouldn’t believe it. Natural as they sound, many chatbot answers are unwieldy, unhelpful and, sometimes, untrue. (These squirrel solutions do seem legit.) Sure, AI can do some of our writing, computer coding and research jobs. But we wouldn’t want to bet our businesses on them alone.
Enter the prompt engineer! This person fine-tunes the prompts that go into a generative-AI large-language model—aka LLM—to extract valuable but buried information for an employer or its clients. Think of it as an AI whisperer.
And there’s a salary premium for the best whispering. On LinkedIn and the job-search site Indeed, thousands of listings came up for the search term “prompt engineer," and among those that stated salary, the annual pay could range from $100,000 to over $200,000. People I spoke with said they were seeing similar pay figures.
“We’ll have to wait to see the long-term effects of gen AI, but it’s clear right now that jobs related to the field are surging," according to a recent report from Indeed.
Hear that? That’s the sound of a whole generation adding “proficient in prompt engineering" to its résumés.
The job is more complex than typing into a text box. To be successful, you need to understand how these systems work, have specific tricks up your sleeve and, in some cases, be able to do some coding. I learned that the hard way after taking a multiweek course then “applying" and “interviewing" for a $200,000-to-$250,000-a-year prompt-engineering role at Hebbia, an AI startup in New York City. (As you’ll see in the video, the company was in on the stunt.)
The job
“You can think of prompt engineering as programming in the English language," George Sivulka, Hebbia’s chief executive and founder, told me. Instead of using computer programming languages like Python, prompt engineers use plain words to achieve results.
Hebbia’s Matrix tool allows financial institutions and other businesses to analyze huge amounts of documents—contracts, filings, earnings transcripts and calls. Instead of an analyst sifting through hundreds of pages to research a company, the AI can summarize and pull out key points for them. Buh-bye, Ctrl-F.
The company’s job posting, the one I responded to, explains what the prompt engineer does: “Your role is to deeply understand our users’ needs, figure out and test the best way to prompt models to meet these needs, and then teach our users."
Translation: Get the robots to cough up valuable answers to keep our human customers happy.
The skills
But what is a great prompt? To find out, I went back to school—and by “to school" I mean “to my couch." Over the past few weeks, I watched portions of Prompt Engineering for ChatGPT, a Coursera class from Vanderbilt University associate computer-science professor Jules White. I paid $49 for the month to get access to the course.
White said I’m one of over 180,000 students who have taken the six-week class in the past year. He said his students tend to be professionals looking to learn more about AI.
LLMs are capable of producing responses with huge variations. The trick is to tweak the “input" to maximize the “output." Here are three approaches learned in White’s class that I now use regularly:
• Persona prompt: Tell the AI system to assume a role. For example: “Act as an assistant helping a financial analyst." Then when you ask your question, the bot will frame the response in that role.
• New-information prompt: You can add new information that the LLM might not know. For example: “Here’s the transcript of Microsoft’s last earnings call. Summarize all the mentions of AI and Bing in five bullets." The more specifically you describe your desired output, the better it will be.
• Question-refinement prompt: You can ask the LLM to suggest improved or alternative questions to get more refined answers. Example: “When I ask a question about Microsoft earnings, suggest a better version of the question and ask me if I would like to use your question instead."
If you want to learn more, I highly suggest you take White’s Coursera course. Just watch it at 1.5X speed. (Sorry, professor!)
During my Hebbia tryout, I came up with some of those Microsoft earnings examples. Swetha Revanur, the company’s head of AI and platform engineering, uploaded several recent earnings transcripts to Hebbia’s Matrix and asked me for some sample prompts. After years of listening to tech earnings calls, I handled that easily. But coding skill is also part of the job, for doing things like building prompts into the system so clients can access them with a quick click. That’s where I fell short.
The future
I won’t be changing my LinkedIn profile to “prompt engineer" anytime soon, but my new skills are already paying off in this job. I often prompt ChatGPT to help me brainstorm and draft emails.
“The future is not artificial intelligence. It’s augmented intelligence with generative AI where we augment and amplify what people can do," White told me. “Everybody, realistically, needs to be interacting with it and using it."
A number of experts in the field told me AI probably won’t get to a point where it understands so much that prompt engineers aren’t needed. “Even in programming where machines are perfect at understanding code, there’s still a job for people to go back and forth," Hebbia’s Sivulka said.
Yet while AI is creating new jobs, it’s also quickly taking other jobs.
The way we communicate with computers has changed a lot in the past year. We now have tools like ChatGPT, Google’s Bard and Microsoft’s Bing, where a few choice words can produce a powerful response. Success in many fields may depend on how well people can sharpen and hone those words.
For the record, I love my job and don’t plan to leave it anytime soon. Unless they hire a prompt engineer to replace me: “You are a tech columnist. Write a column on…"
—Sign up here for the Tech Things With Joanna Stern weekly newsletter. Everything is now a tech thing. Columnist Joanna Stern is your guide, giving analysis and answering your questions about our always-connected world.
Write to Joanna Stern at joanna.stern@wsj.com
3.6 Crore Indians visited in a single day choosing us as India's undisputed platform for General Election Results. Explore the latest updates here!
