Source: Own Illustration
Since the introduction of RWTHgpt at RWTH Aachen University, many people have been asking themselves: How can I use RWTHgpt – and how do I ask the right questions? Because if you want good answers, you have to ask good questions. This is exactly where prompt engineering comes in: the targeted formulation of inputs (prompts) in order to obtain the most helpful, relevant and high-quality results possible from the AI model.
Prompt Engineering – What Does It Actually Mean?
Prompt engineering refers to the deliberate and targeted formulation of input (prompts) in order to obtain the most helpful, precise or creative results possible from an AI model. Language models such as GPT-4 do not “understand” the world like humans do – they work on the basis of probabilities, patterns and statistical correlations. This means that the clearer and more structured a prompt is formulated, the better the model can respond to it.
The Basics of a Good Prompt
An effective prompt is usually more than just a single question. Successful users of GPT models often use a combination of context, instruction and format. The most important elements of a successful prompt are
- Goal clarity: What do I want to achieve – information, text production, analysis, feedback?
- Context: What is the context of the enquiry? What should the model know about the topic, target group or framework?
- Role assignment: If the model “slips into a role”, it can answer more specifically (e.g. as a lecturer, tutor, programmer).
- Format request: In which style or format should the answer appear? List, table, outline, essay, summary?
- Exemplary anchors: Example answers or formulations help the model to orientate itself better.
These elements can be combined as required. Depending on the complexity of the task, the so-called chain-of-Thought prompting: This involves asking the model to “think out loud” or explain intermediate steps. This can be particularly helpful for logical or structural tasks.
What Does This Have to Do With RWTHgpt?
RWTHgpt – the access to various GPT models provided by RWTH – offers a great deal of freedom: Unlike Ritchy, which is only intended for support-related queries, RWTHgpt can be used for a variety of use cases; from academic research to creative ideas for teaching and studying. It offers free input options and can be used for many university contexts: from academic support and learning assistance to brainstorming ideas for events or work processes. In order to utilise this diverse potential, it is important that users know how to ask questions. After all, the effective use of AI requires not only technical understanding, but also didactic expertise.
Application examples from practice
Targeted prompting enables RWTHgpt to be used particularly effectively in numerous application scenarios. These include
- Developing learning tools: Generating exercises, creating multiple-choice questions or preparing content for different levels of knowledge.
- Revising and reflecting on texts: RWTHgpt can act as a “second reader”, suggest alternative wording or provide guidance on ambiguities.
- Structuring ideas: For scientific texts, presentations or concepts, the model can help with the structure and outline.
- Simulate feedback: Teachers can use the model to anticipate typical student questions or present different perspectives.
An example of a precise prompt:
“I am currently preparing a tutorial on “Databases” for first-semester students and am currently writing three explanatory texts on the topic of ‘Primary and foreign keys’ at A2 level. Which figurative comparisons or metaphors would be suitable? Give me an example of each. Your input should serve as linguistic and conceptual support.”
Good Prompting Is a Key to AI Utilisation
Prompt engineering is not a technical gimmick, but a key skill in dealing with generative AI tools. Anyone who wants to use RWTHgpt productively, whether in teaching, studying or administration, benefits from clear, targeted prompts. With RWTHgpt, the IT Center supports a modern, data protection-compliant AI solution for everyday university life. For this technology to realise its potential a basic understanding of communication with AI and therefore good prompting is also required. You can also read more about all our AI services in our previous blog post.
Responsible for the content of this article is Malak Mostafa.
Leave a Reply