As the capabilities of artificial intelligence keep on growing, some companies are hiring 鈥淎I prompt engineers鈥� to help them get the best out of the emerging technology. Are these jobs set to become a ubiquitous presence in workplaces, or are they a passing fad?
Generative AI creates text or images in response to prompts entered by the user. This can be as simple as asking for something in plain English, not dissimilar to using a search engine. But as with search engines, there is a knack to it: including the right references, hints and keywords can guide the AI towards the desired outcome.
Ben Stokes is the founder of , an online marketplace where freelancers sell prompts intended to get the best results from language model chatbots like ChatGPT and image-generating AIs like DALL-E and Midjourney.
Advertisement
鈥淎lthough it鈥檚 just a sentence of text, some people can find it quite hard to approach these generative AI models and put into exact words what they want them to generate,鈥� says Stokes. 鈥淏y buying a pre-built prompt you can ensure it will generate what you want and do it consistently.鈥�
Among the prompts available to buy on PromptBase are templates helping the buyer to create of any species, a prompt that turns ChatGPT into a 鈥溾�� and one that promises to . The average cost is $3.50, but some prompts can cost hundreds of dollars.
Stokes says business is good, with more than 15,000 prompt engineers registered and around four buyers currently to every engineer. Most of the prompt engineers on the site don鈥檛 have a technical background, he says. Engineers specialising in creating images tend to come from a photography or graphic design background, and those working on prompts for generating text often used to work in marketing or copywriting. In essence, they are helping clients use AI to do the same jobs they used to do themselves.
Customers aren鈥檛 always calling in a prompt engineer for their nuance or reliability, says Stokes, but sometimes to cut costs. Companies like OpenAI, the maker of ChatGPT, charge third parties using the service based on the length of the prompt they run, so a carefully crafted request can easily offset its cost.
Specialists on staff
There are signs that companies are keen to hire talent that helps them unlock the benefits of AI. The AI firm Anthropic recently made headlines by advertising a
Aaron Sines at US recruitment firm has been placing engineers into AI jobs for five years, but only came across the term 鈥減rompt engineer鈥� earlier this year. In the second quarter of 2023, his company had 50 prompt engineer roles advertised in sectors from big tech to defence and healthcare, but this has now dropped to 10.
Sines says there is a lot of hype around the term at the moment, and that a lot of the companies approaching him looking for a 鈥減rompt engineer鈥� are actually seeking a more rounded AI researcher with technical skills.
He believes AI models could become better at extracting information from humans about what they want, making the idea of a specialised prompt engineer obsolete, but thinks there will always be a need for humans to oversee AI models during development and application.
鈥淩ight now, salaries for the roles that we have filled are pretty crazy, but the value of these roles is going to be scrutinised and questioned as these systems become more capable,鈥� says Sines.
is a prompt engineer at consulting firm Accenture, having started experimenting with AI in 2021 at a previous job. He is one of around 140 such engineers in his part of the business, mostly made up of existing staff who were trained on AI.
鈥淭his is something new and different because there鈥檚 no instruction manual in terms of how to use these very open-ended capabilities,鈥� says Phelps. 鈥淎 lot of my work isn鈥檛 about building highly proprietary prompts. Really, what I鈥檓 helping my clients with is what are the different prompting strategies that you can use from a technical perspective, and how do you best apply that with your own proprietary data.鈥�
For now, at least, prompts are a valuable type of intellectual property, but one that is practically impossible to protect. at Zhejiang University in China says one of his students purchased several prompts and quickly realised he could have simply re-uploaded them to a different marketplace and started selling them. Qin聽has worked with colleagues to that could 鈥� along with legislation 鈥� help protect the work of prompt engineers.
Qin says complex prompts, and the prompt engineers who create them, are unlikely to disappear any time soon because AI models aren鈥檛 like traditional algorithms that work on logic and can be entirely understood by humans. In many cases, designing a prompt for an AI whose inner workings are mysterious is more like 鈥渕agic voodoo鈥� than programming, he says.
鈥淪ome prompts are short, like maybe 20 or 30 characters. But there are a lot of very, very long prompts. Even the generators of those prompts do not know the exact logic or meanings behind those magical words,鈥� says Qin. 鈥淭he reason is the lack of explainability of the AI algorithm, which leads to the lack of explainability of those prompts. If we cannot solve the explainability problem of AI architecture or algorithms, we cannot solve the problem of these prompts.鈥�
