Communicating Effectively – Crafting Prompts
Far from lines of code and beyond the keywords of search engines, communication with generative AI occurs through a conversational agent, to which a clearly articulated request (i.e., a prompt) must be submitted. The prompt experience is designed to be simple and intuitive. Prompts are intended to establish a clear relationship between the user and the AI.
A computer engineer is no more capable than a novice of generating high-quality content if they cannot clearly convey their expectations to the AI. It is comparable to describing a scene to a visually impaired person: everything depends on how it is formulated. What is not explicitly stated is not considered, and common sense alone is insufficient to bridge the gap between expectations and available information.
The technique of prompting (i.e., prompt engineering) is therefore a skill like any other, involving a number of technical and/or creative considerations. Most importantly, it requires having a clear idea of what one wants to achieve. There are tools available to assist with prompt writing.
Just as it is not necessary to be a mechanic to drive a car, one can use a generative AI tool without mastering all of its technical underpinnings. Nevertheless, to be a professional driver, it is useful to know what is under the hood. The first part of this guide explains the structure of a query.
In the second part, we provide a series of steps and instructions to keep in mind, including specific acronyms as mnemonic aids, to help apply prompting techniques effectively.
S – SPIRIT
Adopt the mindset of a “beginner” or “non-native speaker.” Avoid overly regional or personal expressions, and do not rely on tacit rules or habits.
For example, if you ask a colleague to bring you a “cup of joe,” they would need to understand the slang for “coffee,” locate the coffee maker, know how to use it, and be aware of your preferences (e.g., black with three sugars). If any of this information is missing, they will improvise based on their best guess, which may lead to an unsatisfactory outcome.
Ex.: Explain the CODIR decision regarding the PAT.
Ex.: Explain Decision X of the Executive Committee regarding administrative and technical staff.
T – TRACK
The true strength of these tools lies in their ability to generate thoughtful responses. Ask creative and critical questions to obtain meaningful results and avoid obvious or simplistic answers (e.g., yes/no). To stay on course, it is helpful to have a clear sense of the desired framework and refine it using precise and concise instructions.
Ex.: Is the University Act good?
Ex.: In the context of the 2008 reform of the University Act, have higher education and research been harmonized at the national level?
E – EXECUTION
It is essential to specify the expected action. This can be achieved by using a verb or referring to a specific activity—for example, directive verbs such as “discuss,” “compare,” “design,” or “evaluate.” The choice of verb can significantly influence the outcome.
Ex.: Returning to the University Act example, specifying instructions such as “present the main arguments” or “act as a representative of this political party” will clarify the response and give it a distinct perspective.
P – PARAMETERS
To prevent vague, unprofessional, or content-heavy responses, refine your request by specifying key parameters. For example, indicate a methodology, word count, structure (e.g., email, keywords, report), or format (Python, HTML, CSV, Excel).
Ex.: Write a 450-word paragraph on subject XY in Markdown format.
S – SETTING
Context establishes the environment for the tool. Use limiting terms (e.g., “in Switzerland”), provide familiar examples, or define the target audience. To further shape the response, specify the tone (e.g., formal, casual) or the expected role (e.g., “a quality assurance manager”).
Ex.: Develop a first-year workshop delivered online for adult students.
Ex.: Write a comprehensible response for an 8-year-old child.
Ex.: Write in a formal tone, in the third person, for a university audience.
CONSTRUCTION
There is no single way to communicate with generative AI when submitting queries. It is both normal and important to find your own style, or the one that best suits a given task.
- Once and for all
The approach involves writing a single, precise query. It serves as a starting point to be refined later. This method aims to avoid repeated efforts or successive iterations. It can be applied in various contexts, such as problem-solving, decision-making, or project planning, in order to maximize efficiency and save time and resources.
- Layered
This approach involves submitting a minimal query, evaluating it, and then refining it by adding examples, context, and parameters (see STEPS) until a satisfactory result is achieved. This method also allows a complex problem to be broken down into simpler layers, with each layer responsible for a specific part of the problem. Finally, the tool can be asked to aggregate the results obtained.
- Collage
This approach requires formulating a query that is as complete as possible, rich in context, parameters, and examples, and then iterating through multiple versions of the results. Since the tool generates different outputs for each request, it is possible to take the best parts from each result and combine them to create something new.
It is imperative to adopt an engaged approach. Blindly accepting the results can lead to errors, especially when the task extends beyond the “jagged” boundary where AI capabilities become unpredictable. According to Dell’Acqua et al. (2023), two categories of human-machine collaboration practices have emerged. These distinct strategies were spontaneously adopted in professional settings integrating generative AI. The first delegates tasks either to the tool or to themselves, while the second fully integrates their workflow with the technology.
Centaur Behavior
Like the mythical creature, humans and machines are closely fused in a strategic division of labor. Centaurs discern which tasks are better suited for humans or generative AI based on the strengths and capabilities of each entity, assigning them alternately.
Cyborg Behavior
Cyborgs integrate human and machine capabilities at the level of subtasks. It becomes unclear whether the result was produced by the human or the AI. With this approach, which focuses on complex integration, there is no clear division of labor. Efforts are interwoven up to the boundary of the generative AI’s capabilities.
