Hollywood writers are taking the threat of losing their jobs to AI seriously enough that it’s one of the reasons the WGA went on strike this month. But some creatives who are actively experimenting with tools like OpenAI’s ChatGPT say the sometimes wacky results they get make them wonder if there isn’t an opportunity in finding just the right words to say to finicky machines.
Just as the web and search engines created new digital opportunities for wordsmiths, AI now promises to do the same. Some call the emerging field “prompt engineering” — in other words, twiddling with the textual inputs for AI chatbots until they cooperate and spit out the desired result.
Also Read:
How Hollywood’s Guilds Are Preparing for the Dangers – and Benefits – of AI
A recent Bloomberg article touted “red hot” job listings for “AI whisperers” able to efficiently manipulate prompt words to generate the desired results without too much trial and error. The article cites salaries as high as $355,000 a year, far above the reported average salary for a freelance copywriter of $55,000 a year.
Others are more skeptical: Vice’s tech publication Motherboard called “prompt engineer” a job that “may or may not exist.”
In a recent newsletter, writer-placement agency Gotham Ghostwriters pooh-poohed the notion that prompt engineers are the new wave for writers-for-hire, dismissing the idea as coming from “Redditors with screen names including ‘hornylittlegrandpa’” and concluding that “we’ve seen no sign that elite writers are reinventing themselves as mere technicians.”
Also Read:
This Story Was Not Written by a Robot: AI and the Future of News Media
Still, a website dedicated to job listings in the supposed field called PromptTalent has popped up with more than 20 global job listings. Another job board, Joblist.ai, lists almost three dozen prompt engineering jobs. They’re typically at tech companies, not Hollywood screenwriting gigs, but most that offer pay ranges appear to pay well. One tech company, VidIQ, offers $90,000-$150,000 to write prompts in the service of “developing, designing and maintaining our cutting-edge AI-based conversation systems.”
Story continues
Paul Farnsworth, chief technology officer of tech job listing-site operator Dice, said he’s dazzled by the new landscape of generative AI for both writing and creating images, but so far is seeing prompt writing prowess as more of a plus to mention on your resume, rather than a new career.
“I think it’s a skill right now. It may turn into an engineering discipline, but we’re probably a little bit away from it being a sought-after, highly paid position, to be honest,” he said.
Also Read:
OpenAI Reportedly Lost $540 Million Developing ChatGPT in 2022
The error-prone output of current AI systems certainly seems like it might generate clean-up work after the fact.
Veteran TV writer-producer and first-time novelist Carleton Eastlake (“Monkey Business”) told TheWrap he’s not quite sure what prompt might have saved him when he requested ChatGPT to write his bio. He found his life details and credits seamlessly if inexplicably blended with those of the late Frank Lupo, a writer-producer known for “The A-Team.” Eastlake said in an email that he prefers being himself, though Lupo “made more money.”
Los Angeles novelist and screenwriter Karen Essex also found that ChatGPT needed some assistance of its own when she asked it for help.
Essex recently decided her psychological thriller novel-in-progress would have better “forward momentum” if she switched the action from the past to the present tense, a mindless task that generative AI seemed well-suited for.
“So I started feeding in my chapters with the directive: ‘Rewrite in the present tense,’” Essex, the author of bestselling novel “Leonardo’s Swans,” told TheWrap.
It worked for the first two chapters. Then ChatGPT apparently took “rewrite” literally and liberally, inserting a new, violent first husband into her protagonist’s back story.
It all came down to that word Essex chose in her prompt: “Rewrite.” She discovered her error after the fact at a friend’s baby shower in conversation with the expectant father, Matt Lehman, who serves as head of payments security for Amazon. Only in L.A. — the baby shower conversation turned from the AI-fueled WGA strike to the large-language model powering ChatGPT.
Lehman, whose job makes him conversant with AI, said the prompt word should have been “change,” not “rewrite.” That specific word let ChatGPT go rogue.
Also Read:
The Wacky, Unhinged Bing Chatbot Is Still Good for Microsoft’s Business | PRO Insight
There’s probably not a market in helping writers avoid such mistakes, however. Just as people learned to use search engines in the 1990s and type in just the right terms to find a given result, an individual writer can quickly learn how to tune chatbot queries.
“I suspect that the majority of people are going to learn it pretty quick if you’re interacting with it directly,” Farnsworth said. There’s even AI help for AI help, in the form of AutoGPT, which will ask its robot brethren follow-up questions until it gets the desired result.
It shouldn’t take individual users a long time to learn how to do their own prompt engineering, Lehman said.
More likely, he said, companies may create prompt engineering jobs just to make sure they are perceived to be on AI’s cutting edge.
“There is going to be a huge need to tune models; it’s a viable [job] for a while,” he said.
Companies, he said, “are often going to invest in things that they don’t understand.”
Also Read:
Hollywood, Are You Team Human or Team AI? It’s Time to Choose | PRO Insight
