Artificial intelligence prompt engineering as a new digital competence: Analysis of generative AI technologies such as ChatGPT

Abstract

Objective: The article aims to offer a thorough examination and comprehension of the challenges and prospects connected with artificial intelligence (Al) prompt engineering. Our research aimed to create a theoretical framework that would highlight optimal approaches in the field of Al prompt engineering. Research Design & Methods: This research utilized a narrative and critical literature review and established a conceptual framework derived from existing literature taking into account both academic and practitioner sources. This article should be regarded as a conceptual work that emphasizes the best practices in the domain of Al prompt engineering. Findings: Based on the conducted deep and extensive query of academic and practitioner literature on the subject, as well as professional press and Internet portals, we identified various insights for effective Al prompt engineering. We provide specific prompting strategies. Implications & Recommendations: The study revealed the profound implications of Al prompt engineering across various domains such as entrepreneurship, art, science, and healthcare. We demonstrated how the effective crafting of prompts can significantly enhance the performance of large language models (LLMs), generating more accurate and contextually relevant results. Our findings offer valuable insights for Al practitioners, researchers, educators, and organizations integrating Al into their operations, emphasizing the need to invest time and resources in prompt engineering. Moreover, we contributed the Al PROMPT framework to the field, providing clear and actionable guidelines for text-to-text prompt engineering. Contribution & Value Added: The value of this study lies in its comprehensive exploration of Al prompt engineering as a digital competence. By building upon existing research and prior literature, this study aimed to provide a deeper understanding of the intricacies involved in Al prompt engineering and its role as a digital competence.

Headnote

ABSTRACT

Objective: The article aims to offer a thorough examination and comprehension of the challenges and prospects connected with artificial intelligence (Al) prompt engineering. Our research aimed to create a theoretical framework that would highlight optimal approaches in the field of Al prompt engineering.

Research Design & Methods: This research utilized a narrative and critical literature review and established a conceptual framework derived from existing literature taking into account both academic and practitioner sources. This article should be regarded as a conceptual work that emphasizes the best practices in the domain of Al prompt engineering.

Findings: Based on the conducted deep and extensive query of academic and practitioner literature on the subject, as well as professional press and Internet portals, we identified various insights for effective Al prompt engineering. We provide specific prompting strategies.

Implications & Recommendations: The study revealed the profound implications of Al prompt engineering across various domains such as entrepreneurship, art, science, and healthcare. We demonstrated how the effective crafting of prompts can significantly enhance the performance of large language models (LLMs), generating more accurate and contextually relevant results. Our findings offer valuable insights for Al practitioners, researchers, educators, and organizations integrating Al into their operations, emphasizing the need to invest time and resources in prompt engineering. Moreover, we contributed the Al PROMPT framework to the field, providing clear and actionable guidelines for text-to-text prompt engineering.

Contribution & Value Added: The value of this study lies in its comprehensive exploration of Al prompt engineering as a digital competence. By building upon existing research and prior literature, this study aimed to provide a deeper understanding of the intricacies involved in Al prompt engineering and its role as a digital competence.

Article type: research article

Keywords: artificial intelligence (Al); generative artificial intelligence (GAI); ChatGPT; GPT-3, Bard; Al prompt engineering; digital competences; DigComp; prompting strategies

JEL codes: 033; L26; M15; L86

INTRODUCTION

The artificial intelligence (Al) revolution was propelled by the emergence of unsupervised learning, which enabled models to learn from vast, unstructured datasets (Korzyński, Haenlein, & Rautiainen, 2021). Before 2017, Al focused on narrow tasks through supervised learning, but the introduction of the transformer architecture changed this approach. Models began to identify patterns in data, improving their ability to tackle various tasks. An evolved iteration of OpenAI's Generative Pretrained Transformer (GPT) series, advanced this further by highlighting the potential of prompting models for a better understanding of user context. This emergent property, in-context learning via prompting, has become a fundamental aspect of modern machine-learning models (Cuofano, 2023). Furthermore, Al generative technologies, such as ChatGPT or Bard, have significantly impacted the business landscape, transforming the way organizations operate, communicate, and innovate (Korzyński, Koźmiński, & Baczyńska, 2023; Wach et al., 2023). These technologies facilitate more efficient processes, enhance decision-making, and enable the creation of personalized user experiences, all of which contribute to a more human-centric approach in the era of digital transformation (Korzyński et al., 2023).

Robin Li, the co-founder and CEO of the Chinese Al giant Baidu, predicts that in ten years, half of the world's jobs will involve prompt engineering, and those who cannot write prompts will become obsolete (Smith, 2023). Although this statement might be somewhat overstated, prompt engineers will undoubtedly hold an essential position in the realm of artificial intelligence. These professionals will adeptly steer Al models to generate content that aligns with the desired outcome, ensuring that it is not only pertinent but also cohesive and coherent. Some media indicate that currently, a novel type of Al-related occupation is surfacing, offering six-figure incomes without the necessity for a computer engineering degree or advanced coding abilities. As generative artificial intelligence gains prominence, numerous businesses are seeking to employ 'Al prompt engineers' responsible for refining the performance of Al tools to provide more precise and pertinent answers to questions posed by real individuals. Some of these positions can command annual salaries of up to 335,000 USD (Popli, 2023). Prompt engineering provides remarkable benefits for individuals and organizations working with generative Al models. It allows for greater control over the output, as the right prompts can help ensure Al models create the desired content. Additionally, effective prompts contribute to improved accuracy by guiding Al models to generate more relevant and valuable content. Furthermore, prompt engineering can enhance creativity by presenting Al models with new and unique prompts to explore (God, 2023).

At first glance, one might assume that Al prompt engineering is primarily associated with information technology rather than a human-centric discipline. However, in this article, we aim to offer an alternative perspective. Artificial intelligence prompt engineering is a human language-focused practice in which individuals create prompts using unique and separate elements of text, known as distinct tokens. This method is different from 'soft prompting' techniques, which emphasize automatic finetuning of inputs for machine learning models (Ghazvininejad, Karpukhin, Gor, & Celikyilmaz, 2022). For example, prefix tuning optimizes continuous vectors (numerical representations of data) in pre-trained language models for specific tasks, using 'virtual tokens' instead of distinct ones (Liu et al., 2021). Virtual tokens are continuous, numerical representations of textual elements that can be manipulated and optimized within the model's vector space, offering greater flexibility compared to discrete tokens. On the other hand, prompt tuning applies backpropagation (an algorithm for training neural networks by minimizing the error between the network's predictions and the actual output, allowing the network to learn and improve its performance on various tasks) to train a frozen language model to produce input prompts for particular tasks (Lester, Al-Rfou, & Constant, 2021). In contrast to these approaches, prompt engineering concentrates on manually creating and revising prompts, aligning it more with human-centred fields such as human-computer interaction, and conversational Al, rather than the general field of machine learning (Oppenlaender, Linder, & Silvennoinen, 2023).

This article aims to explore Al prompt engineering as a digital competence using a narrative and critical literature review. We aimed to develop a comprehensive understanding of the intricacies involved in Al prompt engineering and assess its role as a digital competence by examining various dimensions, techniques, challenges, and applications across multiple domains. We asked the following research questions:

1. What are the key dimensions and aspects of Al prompt engineering as a digital competence?

2. What are the technological challenges faced in Al prompt engineering?

3. How is Al prompt engineering applied across various domains such as entrepreneurship, art, and science, and what is its impact in these areas?

4. What are the prominent techniques and strategies used in Al prompt engineering, and how do they contribute to its effectiveness as a digital competence?

The article initiates by outlining the methodological principles tied to the literature review. It then progresses to the crux of the piece, which applies a narrative and critical examination of the literature. Several key topics are delved into, such as digital competencies, the progression of prompt engineering techniques, technological obstacles within prompt engineering, and the utilization of Al prompt engineering across various fields. Subsequently, the article elucidates on Al prompt engineering strategies, founded on the analysis of diverse prompts including text-to-image and text-to-text prompting. It culminates with the final remarks, providing significant implications for educators and organizations.

MATERIAL AND METHODS

In this study, we employed a narrative and critical literature review, following the approach outlined by Ratten (2023), to delve into the various aspects of Al prompt engineering as a digital competence. By building upon existing research and prior literature, we sought to develop a comprehensive understanding of the intricacies involved in Al prompt engineering and its role as a digital competence.

The article systematically explores various dimensions of Al prompt engineering. By analyzing the academic and practitioner literature and conducting desk research, we created a list of potential themes related to Al prompt engineering, including the development of prompt engineering techniques, technological challenges faced in prompt engineering, and the application of Al prompt engineering across various domains like entrepreneurship, art, and science.

Moreover, we developed a conceptual framework that encompasses different strategies for Al prompt engineering. This framework highlights the significance of certain prompt techniques and provides specific examples from business settings.

LITERATURE REVIEW AND THEORY DEVELOPMENT

Digital Competences

The majority of studies on digital competences are related to the education industry, focusing on students' and teachers' abilities to use technology effectively in learning and teaching (Fernández-Batanero, Montenegro-Rueda, Fernández-Cerero, & Garcia-Martinez, 2022; Pozo-Sánchez, López-Belmonte, Rodríguez-García, & López-Núñez, 2020) highlighting the need for continued improvement in this area (Falloon, 2020). These studies indicate the need for digital competences as a core, fundamental literacy addressing both technology mastery and a digital citizenship mindset, especially in higher education (Martzoukou, Fulton, Kostagiolas, & Lavranos, 2020).

Digital competences consist of a blend of knowledge, skills, and attitudes encompassing concepts, facts, skill descriptions (for example, the capability to execute processes), and attitudes (such as a mindset or inclination to take action) (VUORIKARI Rina, Kluzer, & Punie, 2022). Scholars indicate that digital competences transcend mere technical abilities and represent a complex concept that integrates a wide range of skills, including cognitive, emotional, and sociological knowledge, which are crucial for effectively navigating digital environments (Røkenes & Krumsvik, 2016; Wang, Zhang, Wang, & Li, 2021). Moreover, scholars emphasize the importance of critically and reflectively employing technology to generate new insights and knowledge (DeWitt & Alias, 2023; Głodowska, Maciejewski, & Wach, 2023). Over the years, a wide variety of frameworks and models have been developed and utilized to explore and understand digital competence from different perspectives (Blundell, Mukherjee, & Nykvist, 2022; Guitert, Romeu, & Baztán, 2021; Kelentrić, Heiland, & Arstorp, 2017).

One of the most comprehensive developments in this area is the DigComp framework which offers a more in-depth and structured approach to understanding digital competence (Reisoǧlu & Çebi, 2020). Created with the purpose of informing and guiding policies to enhance digital competencies for citizens of all age groups, the framework acknowledges the critical role digital skills play in the constantly changing digital landscape of today (Štaka, Vuković, & Vujović, 2022). The DigComp framework is structured around five core competence areas. Since 2016, these areas include information and data literacy, communication and collaboration, digital content creation, safety, and problem-solving (VUORIKARI Rina eř al., 2022). This framework will be applied to conceptualize Al prompt engineering. We used the DigComp framework to analyze Al prompt engineering as a competence (Table 1).

Al Prompt Engineering

In recent years, Al prompt engineering has emerged as a critical aspect of interacting with large language models (LLMs) such as ChatGPT. By examining the literature, we can identify several themes that help us better understand the applications and implications of Al prompt engineering across various domains. Some of these themes include the development of prompt engineering techniques, the challenges faced by non-experts in prompt engineering, and the application of Al prompt engineering in different domains such as entrepreneurship, art, and science.

Development of Prompt Engineering Techniques

A notable theme in the literature is the ongoing effort to develop and improve prompt engineering techniques. White eř al. (2023) presented a catalogue of prompt engineering patterns that can be applied to solve common problems in LLM conversations. Reynolds and McDonell (2021) introduced the concept of a 'metaprompt,' which is a higher-order prompt designed to seed the language model to generate its own natural language prompts for a range of tasks. The metaprompt approach aims to leverage the language model's inherent capabilities and understanding of natural language to create more effective and nuanced prompts. Instead of directly prompting the model with a specific task, the metaprompt encourages the model to generate appropriate prompts that could be used for the desired task. This approach can potentially lead to more accurate and creative results, as the language model leverages its vast knowledge to generate suitable prompts. Ekin (2023) provides a comprehensive guide on prompt engineering techniques, tips, and best practices for ChatGPT, a powerful tool in natural language processing (NLP) with applications across various industries. This guide covers the fundamentals of prompt engineering, effective prompt crafting techniques, best practices, advanced strategies, and real-world case studies that demonstrate the practical applications of prompt engineering in customer support, content generation, domain-specific knowledge retrieval, and interactive storytelling. Ling et al. (2023) explored the use of reasoning chains as a prompting technique, which leverages structured sequences of related knowledge statements from external knowledge bases to improve the model's performance in open-domain commonsense reasoning tasks.

Moreover, researchers examined zero-shot and few-shot prompting techniques used in natural language processing to enable LLMs to adapt and generalize to new tasks with limited or no supervised training. Zero-shot prompting enables a model to make predictions about previously unseen data without the need for any additional training. This is achieved by framing NLP tasks as natural language prompts and generating responses that indicate the predicted output. However, performance in zeroshot settings often lags behind supervised counterparts, leaving room for potential improvement (Zhou, He, Ma, Berg-Kirkpatrick, & Neubig, 2022). Meanwhile, few-shot prompting is a technique, in which the model receives a small number of examples, typically between two and five, to quickly adapt to new tasks. This approach allows for the creation of more versatile and adaptive text generation models, often with less data required (Meyerson et al., 2023).

Technological Challenges in Prompt Engineering

Another recurring theme is related to technological challenges in designing effective prompts. Bang et al. (2023) found that similarly to other LLMs, ChatGPT suffers from hallucination problems that refer to the phenomenon of LLMs generating plausible-sounding but incorrect or unsupported information in their responses. This issue can be particularly problematic, because it can lead to the dissemination of false or misleading information. However, the interactive feature of ChatGPT enables human collaboration with the underlying LLM to improve its performance through a multi-turn 'prompt engineering' fashion, i.e. an interactive approach in which users iteratively engage with a large language model (LLM) like ChatGPT in a conversational manner. In this approach, users provide input prompts and refine them over multiple turns or iterations based on the responses generated by the model. Li (2023) highlighted the legal and ethical risks associated with LLMs, including stochastic parrots. Stochastic parrot is a term used to describe LLMs' propensity to mimic and reproduce existing text from their training data without a deep understanding of the content. This behaviour can sometimes lead to biased, inappropriate, or irrelevant outputs. Zamfirescu-Pereira, Wong, Hartmann, and Yang (2023) found that non-AI-experts struggled to engage in end-user prompt engineering effectively, often lacking systematic approaches and overgeneralizing their experiences. This difficulty highlights the importance of addressing the challenges related to prompt engineering and the technological limitations of generative Al models, ensuring the safe and responsible use of LLMs across various applications.

Application of Al Prompt Engineering in Different Domains

Entrepreneurship: Short and Short (2023) demonstrated how LLMs can be effectively used to produce entrepreneurial rhetoric by mimicking celebrity CEO archetypes. This study underscores the potential for Al prompt engineering to revolutionize content creation and communication strategies in the world of entrepreneurship.

Art: Oppenlaender, Visuri, Paananen, Linder, and Silvennoinen (2023) investigated prompt engineering as a creative skill for generating Al art, emphasizing that expertise and practice are required to master this skill. This theme showcases the artistic potential of prompt engineering, opening up new avenues for creative expression and collaboration between humans and AL

Science - data extraction: Polak and Morgan (2023) proposed the ChatExtract method using engineered prompts applied to a conversational LLM for accurate data extraction from research articles. This study highlights the potential for Al prompt engineering to streamline data extraction processes and improve the efficiency of research across scientific disciplines.

Healthcare: Wang et al. (2023) provided a comprehensive review of prompt engineering methodologies and applications in the healthcare domain. They highlight the significant contributions of prompt engineering to healthcare NLP applications such as question-answering systems, text summarization, and machine translation.

Al Prompt Engineering Strategies

Text to Image Prompting

Recently, there has been a surge in the number of text-to-image generation models based on LLM (DALL-E2, Stable Diffusion, Midjourney). Noteworthy, the construction of appropriate prompts depends on the tool used at the moment, even though there are general best practices for visual models. Building prompts for text-to-image models is conceptually simpler than for text-to-text models, as the number of tasks is limited in their case.

As mentioned, working with vision models has certain basic principles that we should follow to achieve the desired result. 1) It is important to use a variety of adjectives to give the generated image a desired atmosphere and climate; 2) We should determine the desired image quality, e.g. low resolution, HD or 4K; 3) The style of the image should be specified, such as surrealism, realism, expressionism or photorealism; 4) We should provide a reference name of the artist or creator - such as the name and surname of a painter, graphic designer, or photographer) - to help place the model in the style of a particular work and creator; 5) We should not place more than 2-4 objects in the image. If the number is higher than the generators usually start to become less precise and create blurred content. 6) It is important to be specific as the more specific and detailed the prompt is, the greater the chance of getting the desired result (Oppenlaender, 2022). Table 2 presents the final form of an effective prompt.

Another effective prompting strategy involves learning from existing examples. Resources such as Lexica, a database of images, and the prompts utilized fortheir generation via the Stable Diffusion tool, can be instrumental. Such databases can be navigated to identify desired outcomes or to find inspiration. In the context of brand artwork development or campaign creation, incorporating a consistent descriptive element as a reference point can be highly beneficial. For instance, when crafting artwork for a corporate website, a constant descriptor like 'Bauhaus geometric illustration from 1963' could be applied to each generated image. This approach would yield graphics that adhere to a uniform style, thereby establishing a cohesive visual identity for the website (Parsons, 2023).

Text-to-text Prompting

For models returning text, the creation of prompts usually relies on the desire to solve a specific language task, therefore, the prompt architecture differs from task to task. Table 3 summarizes Al PROMPT engineering recommendations.

Ultimately, an effective prompt may include the following elements: 1) Context - can include information about the role the model should play for the task or any necessary information about the situation that may ground it ('You are a salesperson in a technology company'); 2) Instruction - the task to be performed ('Write an email to a customer thanking for high attendance during the last workshop on MLOps'); 3) Input Data - data for few-shot learning or information the model should use to perform the task ('Customer name: XYZ, keywords to include = [future, innovation, parthership]'); 4) Expected output format - information about the format and type of output in which the answer is to be provided ('Provide just the email message content').

Example 1: 'Tell me what prompt engineering is. Write briefly and without unnecessary details.' This example does not provide a clear structure for the response, nor does it indicate the intended audience. 'Briefly and without unnecessary details' is not a clear instruction. Example 2: 'Write a 3-5 sentence explanation of what prompt engineering is. Write the response as if it were directed to a high school student.' This example provides a clear response structure and specifies the intended audience and thus signals the required tone and vocabulary to use in the response.

Table 4 presents best practices for prompts for frequent tasks that we undertake with the help of LLM. Remember to treat these examples as a good starting point for tests, as each task is individual and needs a specific matching prompt.

In many tasks, we need to combine the following building blocks to receive the expected result. For example, we can use transcripts from the call centre to extract important information and then generate an email based on this information which we will direct to the customer.

The implementation of efficient prompting strategies is essential for improving the performance of LLMs in all possible applications. By following the methods, guidelines, and best practices discussed in this article, users can acquire more accurate, relevant, and fitting responses from both text-to-image and text-to-text generative models.

Resources for Al Prompt Engineering

For the purpose of this article, we analyzed several resources that facilitate the process of prompt engineering in Al systems, such as ChatGPT and DALL-E. These resources range from platforms and databases, like PromptBase and Promptvine, which provide ready-to-use prompts or allow users to exchange their prompts with others. Open-source frameworks, such as OpenPrompt, offer another dimension to these resources. They provide a conducive environment for users to learn, experiment, and even design their own prompts, fostering the development of skills in prompt engineering. Communities also play a pivotal role in the Al and prompt engineering landscape. Platforms like Learn Prompting bring together enthusiasts from various backgrounds to share knowledge, experiences, and best practices. Such communities provide a thriving environment that fosters learning and growth in the Al field. Moreover, there are numerous tutorials and guides, like the Prompting Guide and OpenAI Best Practices, that offer guidance on prompt engineering and general Al practices. These resources serve as invaluable tools, especially for beginners seeking to establish a solid foundation and understanding of Al techniques. Table 5 illustrates resources for Al prompt engineering.

CONCLUSIONS

Artificial intelligence prompt engineering has been playing an increasingly crucial role in various fields, including entrepreneurship, art, science, and healthcare, thanks to its ability to generate targeted and meaningful interactions with LLMs. The DigComp framework, which outlines five core competence areas, i.e. information and data literacy, communication and collaboration, digital content creation, safety, and problem-solving provides a valuable lens to understand and improve Al prompt engineering competencies. The literature reveals a strong focus on the development of prompt engineering techniques, such as the use of metaprompts and reasoning chains, as well as zero-shot and few-shot prompting techniques. These methods are being refined continually to help LLMs adapt to new tasks with limited or no supervised training. However, there are still challenges to be addressed, including hallucination problems and the tendency for LLMs to mimic and reproduce the existing text without a deep understanding of the content. These issues can lead to the generation of false, misleading, biased, or inappropriate information. Furthermore, non-AI experts often struggle with prompt engineering, indicating a need for more accessible and systematic approaches.

Despite these challenges, prompt engineering has shown promising results in a variety of domains. In entrepreneurship, it can mimic celebrity CEO archetypes revolutionizing content creation and communication strategies. In art, it allows for new forms of creative expression. In science, it can streamline data extraction processes from research articles. Finally, in healthcare, it contributes significantly to applications like question-answering systems, text summarization, and machine translation.

In this study, we explored the significant implications of prompt engineering for Al practitioners, researchers, educators, and organizations. We demonstrated how effectively crafted prompts can enhance the outcomes generated by both text-to-image and text-to-text models. Such a skill can lead to the generation of more accurate and contextually relevant results, which in turn improves the overall performance and efficiency of the Al models.

For educators, our findings may serve as a valuable reference for teaching Al students about the role of prompt engineering in optimizing the performance of language models. The strategies and examples we presented could be integrated into curricula to further enrich the learning experience.

In the field of Al research, the strategies and best practices we shared can stimulate the development of more advanced prompt engineering techniques. Our study could also form the foundation for future research examining the impact of prompt engineering on language model performance.

For organizations incorporating Al into their operations, our study underlines the importance of investing time and resources into prompt engineering. By carefully formulating prompts, these entities can derive increased value from their Al systems, as the models will then deliver more precise and helpful outputs.

Last but not least, our exploration yielded a valuable contribution to the field, i.e. the Al PROMPT framework. It provides clear, actionable recommendations for text-to-text prompt engineering, thus helping practitioners refine their prompts to elicit better responses from language models. The Al PROMPT framework emphasizes the importance of articulating instructions, indicating prompt elements, providing ending cues and context, refining instructions to avoid ambiguity, offering feedback and examples, managing interaction, and tracking token length and task complexity.

Sidebar

Received: 24 February 2023 Revised: 22 June 2023 Accepted: 19 July 2023

Suggested citation:

Korzyński, P., Mazurek, G., Krzypkowska, P., & Kurasniski, A. (2023). Artificial intelligence prompt engineering as a new digital competence: Analysis of generative Al technologies such as ChatGPT. Entrepreneurial Business and Economics Review, 11(3), 25-37. https://doi.org/10.15678/EBER.2023.110302

Authors

Authors' contribution was as follows: Paweł Korzyński (40% - conceptualization, draft, revisions), Grzegorz Mazurek (25% - conceptualization, draft, revisions), Pamela Krzypkowska (25% - draft writing, revisions), Artur Kurasinski (10% - draft, revisions).

Paweł Korzyński

Associate Professor at Koźmiński University (Poland). Degree of Habilitated Doctor (dr hab.) in Management (2018), PhD in Economics (2007). Visiting Fellow at Harvard University (2011-2013). Editor-in-Chief of Central European Management Journal (Emerald). His research interests include human resources management, leadership, and technology, including technostress.

Correspondence to: prof. ALK, dr hab. Pawel Korzyński, Koźmiński University, Department of Human Resource Management, ul. Jagiellońska 59, 03-301 Warszawa, Poland, e-mail: [email protected]

ORCID http://orcid.org/0000-0002-6457-4965

Grzegorz Mazurek

Full Professor of Marketing and the Rector of Koźmiński University. In his research he specializes in digital transformation - the new technologies impact on management and marketing. Grzegorz Mazurek authored four books: 'Digital Transformation - marketing perspective,' 'Virtualization of marketing and its impact on network value creation/ 'Blogs and virtual communities in marketing' and 'Promotion on the Internet.' His scientific articles can be found, among others, in Management Decisions, Economic and Business Review, Central European Management Journal, Contemporary Management Research Journal, and International Journal of Cyber Society and Education.

Correspondence to: prof, dr hab. Grzegorz Mazurek, Koźmiński University, Department of Marketing, ul. Jagiellońska 59,03-301 Warszawa, Poland, e-mail: [email protected]

ORCID http://orcid.org/0000-0002-0047-6944

Pamela Krzypkowska

Artificial Intelligence Cloud Solution Architect at Microsoft working with Enterprise market customers on flagship projects in this area in Poland. She focuses mainly on the operationalization of machine learning workloads (MLOps), building machine learning models on Databricks and Azure ML platforms and working with generative models on Azure OpenAI. Pamela is a member of the Al Ethics and Cybersecurity Groups operating next to the Chancellery of the Prime Minister of Poland and a lecturer at the Koźmiński University and the Warsaw University of Technology. She graduated Computer Science (MSc) from the Warsaw University of Technology and Philosophy (BSc) from the University of Warsaw. Her professional interests include neural networks, prompt engineering, generative Al, philosophy of mind, and Al ethics.

Correspondence to: Pamela Krzypkowska, Microsoft Poland, ul AL Jerozolimskie 195a, 02-222 Warszawa, Poland, e-mail: [email protected]

ORCID http://orcid.org/XXX

Artur Kurasiński

Entrepreneur, creator, investor, observer, and critic of phenomena related to technology. He is renowned for his contributions as an author of books, radio dramas, games, and comics. With a tech-realistic approach, Kurasiński is a digital storyteller, curator, and creator. For over two decades, he has been dedicated to observing and deciphering the impact of technology on our lives. As a speaker and participant in Poland's most prominent technological conferences, Kurasiński has established himself as an influential figure in the field.

His professional interests include sociology, media, and storytelling.

Correspondence to: Artur Kurasiński, ul Szpitalna 4 m. 8a, 00-031 Warszawa, Poland, e-mail: [email protected]

ORCID http://orcid.org/XXX

Conflict of Interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

References

REFERENCES

Bang, Y., Cahyawijaya, S., Lee, N., Dai, W., Su, D., Wilie, B., Lovenia, H., Ji, Z., Yu, T.,. Chung, W., Do, Q.V., Xu, Y, fung, P. (2023). A multitask, multilingual, multimodal evaluation of chatgpt on reasoning, hallucination, and interactivity. arXivpreprint arXiv:2302.04023. https://doi.org/10.48550/arXiv.2302.04023

Blundell, C.N., Mukherjee, M., & Nykvist, S. (2022). A scoping review of the application of the SAMR model in research. Computers and Education Open, 3,100093. doi:https://doi.org/10.1016/j.caeo.2022.100093

Cuofano, G. (2023). Prompt Engineering And Why It Matters To The Al Revolution. Retrieved from https://fourweekmba.com/prompt-engineering/on May 10, 2023.

DeWitt, D., & Alias, N. (2023). Creative Digital Pedagogies for Student Engagement: Preparing Students for Industry 4.0. In Digitalization and Development (pp. 112-132). Routledge.

Ekin, S. (2023). Prompt Engineering For ChatGPT: A Quick Guide To Techniques, Tips, And Best Practices. TechRxiv. https://doi.org/10.36227/techrxiv.22683919.v2.

Falloon, G. (2020). From digital literacy to digital competence: the teacher digital competency (TDC) framework. Educational Technology Research and Development, 68, 2449-2472. https://doi.org/10.1007/sll423-020-09767-4

Fernández-Batanero, J.M., Montenegro-Rueda, M., Fernández-Cerero, J., & Garcia-Martinez, I. (2022). Digital competences for teacher professional development. Systematic review. European Journal of Teacher Education, 45(4), 513-531. https://doi.org/10.1080/02619768.2020.1827389

Ghazvininejad, M., Karpukhin, V., Gor, V., & Celikyilmaz, A. (2022). Discourse-Aware Soft Prompting for Text Generation. Paper presented at the Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing.

Głodowska, A., Maciejewski, M., & Wach, К. (2023). Navigating the digital landscape: A conceptual framework for understanding digital entrepreneurship and business transformation. International Entrepreneurship Review, 9(4), Ahead-of-Print.

God, K. (2023). Why Prompt Engineering Is the Job of the Future. Retrieved from https://medium.com/thegenerator/why-prompt-engineering-is-the-job-of-the-future-2a7249b0c82d on May 10, 2023.

Guitert, M., Romeu, T., & Baztán, P. (2021). The digital competence framework for primary and secondary schools in Europe. European Journal of Education, 56(1), 133-149. https://doi.org/10.llll/ejed.12430

Kelentrić, M., Heiland, К., & Arstorp, A.T. (2017). Professional digital competence framework for teachers. The Norwegian Centre for ICT in Education, 134(1), 1-74.

Korzyński, P., Haenlein, M., & Rautiainen, M. (2021). Video Techniques That Help-or Hurt-Crowdfunding Campaigns. Harvard Business Review, 99(2), 29-29. Retrieved from www.hbr.org on May 10, 2023.

Korzyński, P., Koźmiński, A.K., & Baczyńska, A.K. (2023). Navigating leadership challenges with technology: Uncovering the potential of ChatGPT, virtual reality, human capital management systems, robotic process automation, and social media. International Entrepreneurship Review, 9(2). https://doi.org/10.15678/IER.2023.0902.01

Korzyński, P., Mazurek, G., Altmann, A., Ejdys, J., Kazlauskaite, R., Paliszkiewicz, J., Wach, К., & Ziemba, E. (2023). Generative artificial intelligence as a new context for management theories: analysis of ChatGPT. Central European Management Journal, 31(1), 3-13. https://doi.org/10.1108/CEMJ-02-2023-0091

Lester, B., Al-Rfou, R., & Constant, N. (2021). The power of scale for parameter-efficient prompt tuning. arXiv preprint arXiv:2104.08691. https://doi.org/10.48550/arXiv.2104.08691

Li, Z. (2023). The Dark Side of ChatGPT: Legal and Ethical Challenges from Stochastic Parrots and Hallucination. arXivpreprint arXiv:2304.14347. https://doi.org/10.48550/arXiv.2304.14347

Ling, C., Zhang, X., Zhao, X., Wu, Y., Liu, Y., Cheng, W., Chen, H.,Zhao, L. (2023). Knowledge-enhanced Prompt for Opendomain Commonsense Reasoning, working paper. Retrieved from https://charliezhaoyinpeng.github.io/UDMAAAI23/assets/papers/14/CameraReady/Commonsense_Reasoning^_AAAI_UDM.pdf on May 9,2023.

Liu, X., Ji, K., Fu, Y., Tam, W. L., Du, Z., Yang, Z., & Tang, J. (2021). P-tuning v2: Prompt tuning can be comparable to fine-tuning universally across scales and tasks. arXiv preprint arXiv:2110.07602. https://doi.org/10.48550/arXiv.2110.07602

Martzoukou, K., Fulton, C., Kostagiolas, P., & Lavranos, C. (2020). A study of higher education students' selfperceived digital competences for learning and everyday life online participation. Journal of Documentation, 76(6), 1413-1458. https://doi.org/10.1108/JD-03-2020-0041

Meyerson, E., Nelson, M.J., Bradley, H., Moradi, A., Hoover, A.K., & Lehman, J. (2023). Language Model Crossover: Variation through Few-Shot Prompting. arXiv preprint arXiv:2302.12170. https://doi.org/10.48550/arXiv.2302.12170

Microsoft. (2022). CarMax puts customers first with car research tools powered by Azure OpenAI Service. Retrieved from https://customers.microsoft.com/en-us/story/1501304071775762777-carmax-retailerazure-openai-service on July 17, 2023.

Microsoft. (2023a). KPMG augments current capabilities and improves service delivery model with Azure OpenAI Service Retrieved from https://customers.microsoft.com/EN-US/story/1607565448540657318-kpmgprofessional-services-azure-openai-service on July 17, 2023.

Microsoft. (2023b). Strabag SE builds a risk management solution to improve efficiency using Microsoft Intelligent Data Platform. Retrieved from https://customers.microsoft.com/en-gb/story/1596825960615628483-strabagdiscrete-manufacturing-azure-en-austria on July 17, 2023.

Oppenlaender, J. (2022). A Taxonomy of Prompt Modifiers for Text-to-lmage Generation. arXiv preprint arXiv:2204.13988. https://doi.org/10.48550/arXiv.2204.13988

Oppenlaender, J., Linder, R., & Silvennoinen, J. (2023). Prompting Al Art: An Investigation into the Creative Skill of Prompt Engineering. arXivpreprint arXiv:2303.13534. https://doi.org/10.48550/arXiv.2303.13534

Oppenlaender, J., Visuri, A., Paananen, V., Linder, R., & Silvennoinen, J. (2023). Text-to-lmage Generation: Perceptions and Realities. arXivpreprint arXiv:2303.13530. https://doi.org/10.48550/arXiv.2303.13530

Parsons, G. (2023). How to use Al image prompts to generate art using DALL-E. Retrieved from https://create.microsoft.com/en-us/learn/articles/how-to-image-prompts-dall-e-ai on 10 May, 2023.

Polak, M.P., & Morgan, D. (2023). Extracting Accurate Materials Data from Research Papers with Conversational Language Models and Prompt Engineering-Example of ChatGPT. arXiv preprint arXiv:2303.05352. https://doi.org/10.48550/arXiv.2303.05352

Popli, N. (2023). The Al Job That Pays Up to $335K-and You Don't Need a Computer Engineering Background. Retrived from https://time.com/6272103/ai-prompt-engineer-job/on April 15, 2023.

Pozo-Sánchez, S., López-Belmonte, J., Rodríguez-García, A.M., & López-Núñez, J.A. (2020). Teachers' digital competence in using and analytically managing information in flipped learning (Competencia digital docente para el uso y gestión analítica információnál del aprendizaje invertido). Culture and Education, 32(2), 213241. https://doi.org/10.1080/11356405.2020.1741876

Ratten, V. (2023). Research Methodologies for Business Management. Taylor & Francis.

Reisoǧlu, I., & Çebi, A. (2020). How can the digital competences of pre-service teachers be developed? Examining a case study through the lens of DigComp and DigCompEdu. Computers 8i Education, 156, 103940. https://doi.Org/10.1016/j.compedu.2020.103940

Reynolds, L, & McDonell, K. (2021). Prompt programming for large language models: Beyond the few-shot paradigm. Paper presented at the Extended Abstracts of the 2021 CHI Conference on Human Factors in Computing Systems, https://doi.org/10.1145/3560815

Røkenes, F.M., & Krumsvik, RJ. (2016). Prepared to teach ESL with ICT? A study of digital competence in Norwegian teacher education. Computers & Education, 97,1-20. https://doi.Org/10.1016/j.compedu.2016.02.014

Short, C.E., & Short, J.C. (2023). The artificially intelligent entrepreneur: ChatGPT, prompt engineering, and entrepreneurial rhetoric creation. Journal of Business Venturing Insights, 19, https://doi.Org/10.1016/j.jbvi.2023.e00388

Smith, C.S. (2023). Mom, Dad, I Want To Be A Prompt Engineer. Retrieved from https://www.forbes.com/sites/craigsmith/2023/04/05/mom-dad-i-want-to-be-a-promptengineer/?sh=5cf4150d59c8 on April 20, 2023.

Štaka, Z., Vuković, M., & Vujović, V. (2022). The Role of Digital Competencies in the Process of Digital Transformation. Paper presented at the 2022 21st International Symposium INFOTEH-JAHORINA (INFOTEH). https://doi.org/10.1109/INFOTEH53737.2022.9751285

Vuorikari Rina, R., Kluzer, S., & Punie, Y. (2022). DigComp 2.2: The Digital Competence Framework for Citizens-With new examples of knowledge, skills and attitudes. Retrieved from https://publications.jrc.ec.europa.eu/repository/handle/JRC128415 on April 25, 2023.

Wach, К., Duong, C.D., Ejdys, J., Kazlauskaitė, R., Korzyński, P., Mazurek, G., Paliszkiewicz, J., & Ziemba, E. (2023). The dark side of generative artificial intelligence: A critical analysis of controversies and risks of ChatGTP. Entrepreneurial Business and Economics Reviews, 11(2), 7-30. https://doi.org/10.15678/EBER.2023.110201

Wang, J., Shi, E., Yu, S., Wu, Z., Ma, C., Dai, H., Yang, Q., Kang, Y., Wu, J., Hu, H., Yue, C., Zhang, H., Liu, Y., Li, X., Ge, B. Zhu, D. Yuan, Y. Shen, D., Liu, T. & Zhang, S. (2023). Prompt Engineering for Healthcare: Methodologies and Applications. arXivpreprint arXiv:2304.14670. https://doi.org/10.48550/arXiv.2304.14670

Wang, X., Zhang, R., Wang, Z., & Li, T. (2021). How does digital competence preserve university students' psychological well-being during the pandemic? An investigation from self-determined theory. Frontiers in Psychology, 12, 652594. https://doi.org/10.3389/fpsyg.2021.652594

White, J., Fu, Q., Hays, S., Sandborn, M., Olea, C., Gilbert, H., Elnasharx, A., Spencer-Smith, J., & Schmidt, D.C. (2023). A prompt pattern catalog to enhance prompt engineering with chatgpt. arXiv preprint arXiv.2302.11382. https://doi.org/10.48550/arXiv.2302.11382

Zamfirescu-Pereira, J., Wong, R.Y., Hartmann, B., & Yang, Q. (2023). Why Johnny can't prompt: how non-AI experts try (andfail) to design LLM prompts. Paper presented at the Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems, https://doi.org/10.1145/3544548.3581388

Zhou, C., He, J., Ma, X., Berg-Kirkpatrick, T., & Neubig, G. (2022). Prompt consistency for zero-shot task generalization. arXivpreprint arXiv:2205.00049. https://doi.org/10.48550/arXiv.2302.04023