“Even if you went to [ChatGPT creator] OpenAI, they wouldn’t always know what the model was going to produce.”
Some prompt engineers cannot code and solely rely on writing clear and precise text-based prompts to achieve a desirable outcome. KPMG calls these workers “prompt crafters”.
Others have software engineering backgrounds and, in addition to crafting prompts, are expected to tweak the back-end code that powers a given AI tool to make it more likely to produce what a company wants.
For Damm, who falls in the second category, the unpredictability of generative AI means much of prompt engineering comes down to trial and error.
His team reads “heaps of literature” to stay on top of the best prompting techniques.
For example, a 2023 study by Chinese academics, which Damm did not cite, found that inputting phrases such as “this is very important to my career” enhanced the performance of various large language models such as ChatGPT and GPT-4.
Months later, a ChatGPT user shared on social media platform X that the generative AI tool gave longer responses to questions when offered a tip.
Some research suggests ChatGPT delivers better results when users tell the tool their query is important to their career. AP
“Those quirks are going to change over time as the models change,” Damm says.
“And I think that’s why there’s going to be a real need for us to have an army of people who are capable of leveraging what the AI [tools] are able to do.”
Amid warnings that generative AI could destroy millions of jobs, the emergence of prompt engineers offers an insight into the type of job the technology might create.
Some believe the occupation has a short shelf life and will be phased out when generative AI becomes advanced enough to understand our requests without the need for these “AI whisperers” to act as intermediaries.
But others, including the tech leaders at KPMG, appear more bullish about prompt engineering’s prospects.
Executives at the firm see enough value in the discipline to employ four full-time prompt engineers, a contingent that includes Damm, and they want to triple that number by the end of the year to meet rising internal demand for AI-powered solutions.
Other KPMG employees use prompt engineering in their roles but have different job titles. About a quarter of the firm’s workforce has undertaken basic prompt-crafting training, too.
“We’re starting to build a library [of effective prompts], and we’re really keen to see how we can leverage what we learn over time to be reusable,” Damm says, adding that KPMG was committed to making prompt engineering a “discipline”.
“To make it a discipline, [it] isn’t just [about] having individuals who are prompt whisperers. It’s [about] being able to bring what they learn … to a central repository that we can then leverage to train others to do the same.”
Job ads virtually non-existent
Data from employment site Seek suggests the occupation of prompt engineering is in its infancy in Australia.
On May 30, just two active job ads on the platform contained the word “prompt” in their title, while just five ads contained “prompt engineer” in the details section. Ten jobs contained the phrase “LLM”, which refers to large language model, and 67 contained the phrase “generative AI”.
BOSS asked several major employers whether they employed prompt engineers.
NAB said it did and had also upskilled people across the business. CBA said it was considering carving out some prompt engineering roles, but currently this work was done by software engineers.
Meanwhile, PwC, ANZ, Deloitte, Accenture and WiseTech Global said none of their employees had the title “prompt engineer”, but many did prompt engineering as part of their roles and many employees had been trained in how to write effective prompts.
Matthew Munson, managing director of technology recruitment business Talent, says the occupation of prompt engineering is more widespread in the US than in Australia, partly due to a lack of investment in local startups during the past 18 months.
He tells BOSS he hasn’t yet been asked to recruit a non-coding prompt engineer (or a “prompt crafter”, to use KPMG’s terminology), but prompt engineering roles that involved coding typically paid between $170,000 and $180,000 a year plus superannuation.
“Despite the lack of prompt engineering roles in Australia currently, the impact of LLMs and AI will be like the birth of the internet,” Munson says. “Prompting is currently a tool for software engineers but in time will grow across all roles and industries.”
‘Like boasting you can use a search engine’
Others are less convinced that the job title will stick around.
Spirit Technology Solutions non-executive director Lynn Warneke, who is also a member of the Australian Computer Society’s AI Ethics Committee, says prompt engineers are unlikely to become a common job title as the skill of prompt engineering will soon be “democratised”.
“But we might see a kind of skill set emerging around really expert, higher-order [prompt engineering] skills,” Warneke says.
QUT sessional academic Cameron Shackell, who holds a PhD in Semiotics and Information Technology, is even more sceptical.
“Eventually, boasting that you know how to prompt AI will become resume furniture,” Shackell wrote in an article for The Conversation.
“It will be comparable to boasting you know how to use a search engine (which wasn’t always so intuitive) – and may paint you as a dinosaur if mentioned.”
