SYLVIE DOUGLIS, BYLINE: NPR.
(SOUNDBITE OF DROP ELECTRIC SONG, "WAKING UP TO THE FIRE")
ADRIAN MA, HOST:
It's Jobs Friday.
(SOUNDBITE OF AIRHORN)
MA: The best time of month for the horn of the air variety.
DARIAN WOODS, HOST:
I never thought my life would involve so much cheering over the air horns, but here we are.
MA: (Laughter). That's not what you thought economics was going to be all about, but, you know, it's how we roll. Well, you know, Jobs Friday, of course, is this time of the month where we look at how the economy is working for real people and spotlight trends in the job market.
WOODS: Yeah. According to the latest jobs report, the economy added 206,000 jobs in June, and the unemployment rate ticked up from 4% to 4.1%. It's the first time it's reached that level in close to three years. So overall, the numbers suggest a modest cooldown in the labor market.
MA: Although, this Friday, we're focusing on a job that does not seem to be cooling down. And it's one that I'm willing to bet many of you have never even heard of - prompt engineer, specifically AI prompt engineer.
WOODS: Nick Bunker is an economist for the jobs listing site Indeed. He's also a friend of the show. And Nick says posts calling for prompt engineers didn't seem to even exist until very recently.
NICK BUNKER: So when I look back, it does not register in any real material way until late fall, early winter of 2022.
MA: And yet, Nick says demand for prompt engineers has increased about tenfold in the past year.
BUNKER: Yes, it is a very rapid growth. There's still very, very few of these jobs, but they've grown very, very rapidly. And this is something we actually see in the context of all generative AI job postings.
WOODS: Prompt engineers are in demand with some employers advertising six-figure salaries. But what exactly do they do?
(SOUNDBITE OF MUSIC)
WOODS: This is THE INDICATOR FROM PLANET MONEY. I'm Darian Woods.
MA: And I'm Adrian Ma. Today on the show, we find out by talking to a man who became a prompt engineer pretty much by accident.
(SOUNDBITE OF MUSIC)
MA: Sometimes, life throws you a curveball. Back in 2020, life threw Stephen Webster a curveball, a change-up and a slider all at the same time.
WOODS: Stephen was living in Austin, Texas, and working as the executive editor of austin.com, a website that promotes tourism and events around the city.
STEPHEN WEBSTER: During the pandemic, when all the public events went away, so did our revenues. Shortly after the business collapsed, I became a dad. It was one of those moments in life where you kind of have to rethink everything. And as somebody with a writing background, for me, it was pivoting into tech as a technical writer.
MA: So Stephen got a contractor gig with Google, which focused on writing how-to guides and product instructions. That was back in early 2022. Then several months into this role, a company called OpenAI released something called ChatGPT.
WOODS: Never heard of it.
MA: (Laughter) Yeah. So, as we all know, ChatGPT ignited this cultural obsession we all currently have with AI and all the ways it could change our personal and work lives.
WOODS: Stephen says Google saw this happening and was like, we have to step things up.
WEBSTER: Google was, all of a sudden, playing catch up. So they, like, requisitioned all of their technical writers, and suddenly, we were forced to become prompt engineers.
WOODS: Now, Stephen hadn't really planned on another career pivot. But this turned out to be a lucky accident for Stephen. He had already taken a casual interest in artificial intelligence, and prompt engineering became his foothold in the industry. For the past couple of years, it's been his main job.
MA: So what exactly does a prompt engineer do? Well, a prompt is basically the question or instruction that you give to an AI. That AI could be based on a large language model like ChatGPT is or a website like DALL-E, which lets you generate images just by typing in words. And when you do that, you are giving prompts. But getting those answers to the point they're actually usable and not just a bunch of incoherent garbage - that takes a lot of tweaking on the back end.
WOODS: And that is where prompt engineers come in. They help develop an AI app or an AI model by testing and tweaking, and, of course, prompting it again and again and again until the AI actually gives you a useful answer without bugging out. Sounds pretty simple, right?
(SOUNDBITE OF TV SHOW, "THE DAILY SHOW")
JON STEWART: Prompt engineer? I think you mean types question guy.
(LAUGHTER)
STEWART: (As self) And by the way...
WEBSTER: One of my heroes, Jon Stewart, the host of "The Daily Show" - he called prompt engineers the asks questions guy, which I thought was really clever.
(SOUNDBITE OF TV SHOW, "THE DAILY SHOW")
STEWART: ...Engineers. And by the way, you're not fooling anybody by adding the word engineer. You're not the types question guy, you're the vice president of question input.
WOODS: It helps not to have heroes. They'll disappoint you.
MA: (Laughter) Yeah. I mean, Stephen is a pretty good sport about this, actually. But he says there is a little more to the job than just typing questions. Anyone can prompt an AI. But he says making sure the answers are actually useful takes creativity and experimentation.
WOODS: So Stephen gives an example from his personal life. Recently, he wanted to create an AI cooking assistant using ChatGPT. This will be one that he could talk to that could do meal planning and give him tips in the kitchen. And in an earlier version, he trained this AI bot by feeding it transcripts of YouTube videos that featured various famous chefs. This, however, had some unintended consequences.
WEBSTER: I asked it a question that it considered to be stupid. I think it was like, how long should I cook these noodles or something like that? And it's like, listen, you stupid doughnut, you cook it until you like its doneness, OK?
WOODS: Wow, a dominant chatbot.
MA: Rude. And in case you're wondering where that doughnut insult comes from, Stephen thinks the AI learned it from a video of the famously foul-mouthed celebrity chef Gordon Ramsey.
(SOUNDBITE OF TV SHOW, "HELL'S KITCHEN")
GORDAN RAMSEY: (As self) Taste that. It's a block of parmesan, you [expletive] doughnut.
WOODS: Yes, not the friendly kitchen assistant that Stephen was looking for. So he went back to the drawing board. And one of the things he did was prompt the AI by giving it a friendlier persona and a more detailed backstory. He told it, you are Chef Julia.
WEBSTER: Chef Julia used to work in high hospitality at a three-Michelin-starred restaurant, and she quit. And so she opened up an Irish pub, and she's now well known for making Scotch eggs, which is her favorite recipe...
WOODS: Wow.
WEBSTER: ...From her youth. And providing that level of designation to an AI agent enhances its logical performance by 10 to 20% over the stock AI.
MA: After about 300 or 400 rounds of prompting and revising and tweaking, Stephen has gotten Chef Julia to the point where I could actually talk to her on my smartphone using the ChatGPT app.
AI-GENERATED VOICE: (As Chef Julia) Do you have any particular types of cuisine or flavors you enjoy?
MA: Chinese.
AI-GENERATED VOICE: (As Chef Julia) Chinese cuisine is fantastic. For your first meal, how about a classic stir fry?
BUNKER: It's sounding a little obsequious to you, Adrian.
MA: Oh, well, then it called me a stupid doughnut.
WOODS: Oh, right. Well...
MA: No, I'm kidding. It didn't do that.
WOODS: (Laughter) Stephen says this is what the job of prompt engineering is about, making AI that is actually useful, which is why companies want to hire these people.
MA: Now, after Stephen's early foray into prompt engineering with Google, he moved on to other companies. Currently, he's working at a staffing and marketing firm called Aquent, and one of their clients is a pharmaceutical company. Stephen says he's helping build this AI tool that, among other things, can generate articles complete with citations to scientific research. And while he says prompt engineering does seem like a hot job right now, he can already see a future where AI becomes so advanced that prompt engineers become obsolete.
WOODS: Yeah, he reckons that AI will eventually be able to prompt and tweak itself.
WEBSTER: I don't think that prompt engineering is going to be around in 10 years. I do think that prompting AI is going to be a skill that literally everybody will learn, kind of like Microsoft Office was or social media marketing. You know, it'll become a daily part of your life as AI assistants. And so prompting is simply what you do when you interact with that AI.
MA: And what about your own job? How do you feel about that?
WEBSTER: I feel just fine. I don't intend to stay here forever, and I don't think anybody who gets into prompt engineering should because as AI evolves, we're going to have to evolve with it.
(SOUNDBITE OF MUSIC)
MA: Yes. We love a pivot king.
WOODS: He is the pivot master.
MA: This episode was produced by Cooper Katz McKim, with engineering by Kwesi Lee. It was fact-checked by Sierra Juarez. Kate Concannon edits the show, and THE INDICATOR's a production of NPR.
Copyright © 2024 NPR. All rights reserved. Visit our website terms of use and permissions pages at www.npr.org for further information.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.