The widespread public use of tools that employ generative artificial intelligence for informative and creative texts, translations and information, images and videos has made it necessary to learn how to communicate effectively with “the machine” through prompts.
Why has the role of prompt engineer emerged?
While some sectors of public opinion and many professional fields are concerned that the advent of artificial intelligence may lead to the disappearance of many types of jobs in the near future, it is also recognised that AI is creating a favourable context for the emergence of new professional roles [14].
One of these new roles is the prompt engineer [16]. To understand this position, it is important to note that with the development of large language models based on machine learning, and their widespread adoption by the general public, the way in which users formulate their requests verbally (the so-called “prompt”) during interactions with the system has become particularly significant.
These models are based on a probability distribution of the terms in a given language, which is calibrated during training using a set of documents supplied to the machine as examples of how those words are used in context.
In this way, the system produces as output the result it calculates to be most likely given the particular prompt it receives, thereby ensuring the plausibility and coherence of the response, but not necessarily its truthfulness in relation to the data on which the learning process was based.
The need for prompts was neither anticipated nor specifically developed as a feature. This led, following the release of OpenAI's ChatGPT at the end of 2022, to the emergence of “special instructors”, prompt engineers, whose role is to teach an artificial intelligence algorithm how to interpret human language in order to produce desired, and above all, useful results.
Who is the prompt engineer?
A prompt engineer, in any business or professional sector, is someone dedicated to crafting and inputting prompts, rules, and instructions designed to make the most of new generative artificial intelligence services. Generally, this is a person with programming experience who works with large language models (LLMs) such as those behind ChatGPT and Bard (text-to-text prompting), or DALL-E 3 (text-to-image prompting).
They must also have strong skills in data processing and information extraction, as well as significant experience with software tools for creating digital content [11]. In addition to these general competences, which apply regardless of the field in which AI is being used, a prompt engineer is expected to have an in-depth understanding of both language and subject matter, in order to adapt general prompt creation guidelines to specific contexts. For example, they need to be able to generate accurate and persuasive marketing and communication content, and critically assess the outputs provided by LLMs to identify incorrect or unsupported responses (known as hallucinations).
Since it is difficult to find all the varied skills described above in a single individual, it is reasonable to consider forming teams of prompt engineers, in which the knowledge and expertise of each member is combined. According to some opinions in the literature [10], this scenario also represents an interesting opportunity for those nearing the end of their careers, who possess a wealth of experience in their field but may lack familiarity with the latest developments in artificial intelligence.
What does a prompt engineer do?
AI language and natural language skills are essential for prompt engineers, as they enable them to provide advanced input texts, specific labels, or to devise input strategies to train and guide the language model towards producing outputs tailored to a very specific goal. Prompt engineers work on natural language processing (NLP) projects and are responsible for creating input texts (the prompts) according to the outputs required by the business and the project, with the aim of improving the performance of language models.
For example, an online food retailer might have thousands of images of fruit and vegetable products, but none of the image metadata describes which fruits and vegetables are present in which photos. Asking the language model to display “an image containing red peppers” is linguistically different from asking for “a photo of red peppers.” In the first case, you might receive an image of a red pepper salad; in the second, a picture of red peppers. The way a prompt is written determines the AI model’s response, and thus both the effectiveness of the prompt and the outcome you wish to achieve.
Which companies need prompt engineers?
Currently, large companies are seeking prompt engineers to optimise the competitive advantage offered by generative AI. However, SMEs and start-ups across all sectors—from healthcare to retail—will increasingly require prompt engineers with expertise in programming, machine learning and deep learning principles and techniques, statistical and mathematical concepts, and natural language processing. In practice, across every sector—from communication to marketing, industry to sales—prompt engineers are professionals who possess specialist expertise relevant to the business area in which they work, such as medical communication, art, law, marketing, CRM, and so on.
Countless recent practical examples demonstrate the sharp rise in interest in this professional role among companies in a wide range of sectors. To mention a few striking cases: Anthropic, a start-up funded by giants like Google [7] and Amazon [4], has offered a salary of $335,000; Klarity, specialising in automated document review software, has offered $230,000; and the consultancy firm Booz Allen Hamilton, $212,000 [8]. In the healthcare sector, one of the first hospitals to begin searching for prompt engineers was the prestigious Boston Children’s Hospital [3], which ranks among the top hospitals in the United States [6].
It should be pointed out, for completeness, that according to some views [5], the enthusiasm of companies for this type of role may prove to be temporary, as future improvements in interaction mechanisms with LLMs could render the job redundant [17]. Alternatively, it might represent an economic bubble linked to the initial phase of adopting a new technology (Gartner Hype Cycle), which creates expectations that are objectively higher than can be justified by the current level of development [18][19].
Ultimately, as for the future, it is important to note that the ability to communicate with generative artificial intelligence systems using natural language has only very recently become accessible to the general public (the key moment being, as mentioned, the launch of ChatGPT on 30 November 2022). As a result, it remains quite difficult to predict how the phenomenon and related factors, such as prompt engineer employment, will evolve [16].
References
[1] https://www.wired.it/article/intelligenza-artificiale-come-diventare-prompt-engineer/
[2] https://www.intelligenzaartificialeitalia.net/post/cos-%C3%A8-la-prompt-engineering-o-ingegneria-dei-prompt
[3] https://mashable.com/article/ai-healthcare-integration
[4] https://www.wired.it/article/amazon-investe-4-miliardi-startup-intelligenza-artificiale-anthropic/
[5] https://mashable.com/article/what-are-prompt-engineer-jobs-ai
[6] https://health.usnews.com/best-hospitals/area/ma/boston-childrens-hospital-6140270
[7] https://www.reuters.com/markets/deals/alphabet-backed-ai-startup-anthropic-raises-450-million-funding-freeze-thaws-2023-05-23
[8] https://time.com/6272103/ai-prompt-engineer-job/
[9] https://mashable.com/article/ai-new-jobs
[10] K. Chetty. AI literacy for an ageing workforce: leveraging the experience of older workers. OBM Geriatrics, 7(3), 1-17, 2023.
[11] P. Korzynski, G. Mazurek, P. Krzypkowska, and A. Kurasinski. Artificial intelligence prompt engineering as a new digital competence: analysis of generative AI technologies such as ChatGPT. Artificial intelligence, 11(3), 2023.
[12] B. Meskó. Prompt engineering as an important emerging skill for medical professionals: tutorial. Journal of Medical Internet Research, 25, 2023.
[13] J. G. Meyer1, R.J. Urbanowicz, P.C.N. Martin, K. O’Connor, R. Li, P.‑C. Peng,
T.J. Bright, N. Tatonetti, K.J. Won, G. Gonzalez‑Hernandez, and J.H. Moore.ChatGPT and large language models in academia: opportunities and challenges. BioData Mining, 16, 2023.
[14] T. Orchard, and L. Tasiemski. The rise of Generative AI and possible effects on the economy. Economics and Business Review, 9(2), 9-26, 2023.
[15] J.D. Zamfirescu-Pereira, R.Y. Wong, B. Hartmann, and Q. Yang. Why Johnny can’t prompt: how non-AI experts try (and fail) to design LLM prompts. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems, pages. 1-21, 2023.
[16] T. Teubner, C.M. Flath, C. Weinhardt, W. van der Aalst, and O. Hinz. Welcome to the era of ChatGPT et al. The prospects of large language models. Business & Information Systems Engineering, 65(2), 95-101, 2023.
[17] O.A. Acar. AI Prompt Engineering Isn’t the Future. Harvard Business Review, 2023.
[18] https://www.gartner.com/en/articles/what-s-new-in-artificial-intelligence-from-the-2023-gartner-hype-cycle?_its
[19] https://totalent.eu/generative-ai-on-the-garner-hype-cycle-is-the-ai-hype-finally-dying-down
