Introduction

Artificial Intelligence (AI), and specifically Large Language Models (LLMs), are dynamically entering every aspect of professional and daily life. The ability to communicate effectively with these advanced systems, a skill known as “prompt engineering,” is emerging as a fundamental digital literacy.

This guide was curated by leading Greek AI scientists to make this new technology accessible and understandable to the general public. It is addressed to every citizen, employee, and public or private sector entity wishing to utilize AI in everyday life and at work, without requiring prior technical knowledge. Through clear examples and useful practices, it presents communication methods that improve the quality, clarity, and usefulness of LLM responses.

The central goal is user empowerment: it explains how to define the system’s role, formulate clear instructions, describe the target audience, and request an appropriate response format. At the same time, it highlights the limitations of LLMs, such as “hallucinations,” data biases, and uncertainty, to cultivate a critical mindset and apply practices for verifying sources and cross-checking information.

The guide aims to spread safe and beneficial ways of utilizing AI across a wide range of tasks. Through examples, it offers a practical reference point that helps everyone use innovation responsibly, maximizing benefits and limiting risks. In this way, digital trust is strengthened and the diffusion of AI for the benefit of society is supported.

Artificial Intelligence (AI) is a rapidly growing branch of computer science with a strong interdisciplinary character, as it draws knowledge from many disciplines, including mathematics, biology, linguistics, and economics. The most recent AI achievement is Large Language Models (LLMs), which significantly affect many aspects of daily and professional life. This section aims to clarify the basic concepts of AI and LLMs to understand their operation and capabilities.

AI focuses on developing systems capable of operating with varying levels of autonomy to perform complex tasks, such as translation or driving vehicles. Like all computer systems, their operation is based on three stages:

Modern AI systems are not programmed with detailed instructions for every step. Instead, they are “trained” by processing a massive volume of example data. For instance, a transcription system is first fed millions of audio recordings paired with their corresponding transcriptions. Through this process, the system gradually adjusts its operation to achieve specific, measurable, and well-defined goals, which in this case means converting speech to text with the highest possible accuracy.

Thus, the “training” process of modern AI systems requires two key elements:

Modern AI systems help solve highly complex scientific problems, such as predicting the 3D structure of proteins (vital for designing new drugs), rapidly analyzing medical images for early diagnosis, or modeling the impacts of climate change. This contribution is so critical that it has supported research leading to recent Nobel Prizes.

LLMs are a type of AI system that has brought significant changes to how users interact with technology in recent years. LLMs are trained on massive collections of text (e.g., websites, books). In the first stage, they learn to predict which word is most likely to follow a given piece of text. For example, if given an unfinished story, they can complete it by adding the most probable next words one by one.

This capability also allows LLMs to answer questions. If they receive a question as input (e.g., “What is the capital of Greece?”), a highly probable continuation of the text is the correct answer (e.g., “Athens”). To make responses more accurate or appropriate, modern LLMs are fed thousands of instruction examples (e.g., “Write a recipe”) alongside the ideal, desired answers. At this stage, they also receive a positive “reward” when they provide polite answers and when they refuse to execute malicious requests. More recent LLMs have also been trained on multimodal data (images, speech), enabling them to respond to spoken requests or generate images.

LLMs rely on “neural network” technology, a type of AI inspired by how the human (and animal) brain is organized, although it does not function in the exact same way. The adjective “large” in the term “Large Language Models” refers to the fact that modern LLMs possess billions of artificial neurons that exchange information with each other. The connections between these neurons are adjusted during training.

Because LLMs have “read” vast amounts of data (websites, books, etc.), they can simulate many different roles, like answering as an expert or an ordinary citizen, or writing in a specific tone. That is why it is important for the user to specify in their requests exactly how they want the answer to be generated.

These requests are called prompts. A good prompt usually includes the role the system should play (e.g., “you are an experienced technical analyst”) and specifies the desired format of the response (e.g., how short or detailed the output should be, and what tone to use). Through appropriate prompting, impressive results can be achieved across a variety of useful processes. This is why this guide provides tips on writing effective prompts.

The emergence of LLMs is changing how users interact with digital information. Interacting with LLMs via prompts differs from traditional internet searching. Unlike the simple keywords used in search engines, an LLM prompt can be a straightforward question (“What is the capital of Greece?”), which the system will process. However, it can also define a role for the LLM to simulate, provide relevant examples, or give instructions on the desired response format. The format and clarity of the prompt directly impact the quality and usefulness of the output generated by the LLM.

The main differences between traditional search and LLM prompting are summarized below:

Many LLMs now offer the ability to accompany their answers with citations pointing to the webpages or other information sources they used. This allows for the verification of LLM responses, which is a critical step for their reliability.

At the same time, there is a gradual integration of LLMs into search engines, fundamentally transforming the information retrieval experience. However, LLM users should always verify their answers, because these systems often generate highly convincing text, images, and speech, which may nevertheless contain incorrect information (known as “hallucinations”).

LLMs can significantly boost human productivity and creativity. As an example, some of the most well-known LLMs available to the general public include OpenAI’s ChatGPT, Anthropic’s Claude, Microsoft’s Copilot, Google’s Gemini, and Meta’s LLaMA (for more information, we refer the reader to the Appendix).

Their capabilities range from writing and editing text to data analysis, programming, translation, and content creation. LLMs have also been integrated into many everyday applications. Examples and techniques for effective prompt writing are provided in the following chapters.