Generative AI
What is it, and how can you use it in your studies?
Here on this page, you can get a brief overview of what the "new" generative AI is, how to work effectively with chatbots, and what you should particularly pay attention to when using it in connection with your studies at KEA.
What is generative AI?
How does the technology work?
When we talk about AI, it’s surrounded by mysterious words and abbreviations: Chatbot, LLM, ML, NLP, just to mention a few you may have encountered. Countless books and articles have been written about each topic, but our focus here is not so much on the underlying technology, but rather on the use case in an academic context. However, it makes sense to briefly touch on the mentioned concepts, as they are fundamental to understanding what it really means to work with generative AI.
LLM stands for large language model and can be compared to the neural networks that our brains consist of. They contain millions of "brain cells" trained on large amounts of data using machine learning (ML). Language models have existed in various forms for many years but had a major technological breakthrough in 2017, which laid the foundation for the generative AI wave we’re currently experiencing. For years, we’ve been exposed to chatbots from businesses and public institutions, which, without much success, have tried to simulate human dialogue to save employee resources. With simple text interfaces like OpenAI's ChatGPT or Google’s Gemini, it’s now finally possible to create something that resembles a real conversation with a computer. This is possible through natural language processing (NLP), which, in short, enables computer programs or chatbots to understand and even generate text and speech. It’s important to understand that the responses you receive are simply expressions of statistical probability based on the data and training the language model has received.
In simplified terms, you can say that chatbots are just probability machines, providing the most likely output for a given input. In other words, there’s no critical thinking or understanding of the context or world we live in. This is an important point to keep in mind when working with generative AI in your studies!
Remember your critical thinking!
Bias, hallucinations, sensitive data, and copyright
When you use AI in an academic context, there are several things it is important to be aware of. Here, we will briefly touch on four of the most important: bias, hallucinations, sensitive data, and copyright.
Bias and prejudices are always part of the data foundation
First of all, it is obvious that the data on which language models are trained is not neutral, but contains all the same biases or prejudices as any other material. For example, if you trained two versions of the same language model on different datasets from a conservative party and a left wing party, it is clear that you would get different answers if you subsequently prompted them individually about what the optimal distribution policy in Denmark would be. The various chatbot developers try hard to balance and eliminate bias, but realistically, it is an impossible task to create a completely neutral chatbot. Therefore, it is very important to always remain critical and assess whether there might be hidden biases in the form of prejudices, cultural differences, or lack of representation of viewpoints.
AI chatbots are not sources of factual truth
Another point of attention is the tendency of generative AI to generate misleading or outright false information – so-called hallucinations. Here, we are not dealing with bias or lack of representation in the data foundation, but simply pure fabrication. Remember that AI chatbots are just probability machines, constantly generating the most probable next word in a sentence based on training data and prompts, without ANY form of reflection, fact-checking/validation, or critical thinking. It can seem very convincing since chatbots are designed to create polished texts, but precisely for that reason, it is extremely important not to be swayed by eloquence but ALWAYS double-check and verify sources and information.
Think about what data you share
When it comes to what data you provide to a chatbot, you should always be cautious. As a rule, never share personal data, as it may be stored. Instead, use anonymized data, and be especially mindful of whether there are things that should not be shared with others, such as when copying/pasting large amounts of text or uploading documents.
Do not upload anything you did not write yourself
Finally, copyright is a rather complex topic with many gray areas. As in many other areas of the digital world, legislation struggles to keep up, but as a general rule, it is a good guideline not to upload anything you did not write yourself.
Prompting
How to use chatbots
The internet is flooded with guides on prompting and "how to become a prompt engineer" courses. The word "prompt" has taken the world by storm, and what was once a niche term reserved for coders and tech enthusiasts is now everywhere. It can be debated how much sense all the focus and hype around prompts make, both because there are so many different language models trained differently and continuously changing their functionality, but also because there’s a certain amount of superstition and magical thinking involved, where anecdotal successes are elevated to methodological truths. Here, we’ll attempt to keep it brief with a few tips on how to prompt chatbots more effectively, but keep in mind that this should be taken with a grain of salt and is constantly changing as the technologies evolve.
There may be two main schools of thought on prompting, with the first working with so-called "megaprompts," which try to include everything in one large prompt. The hope is that you’ve constructed your prompt with all the necessary information the chatbot needs to provide a complete answer. In theory, one of the advantages of using megaprompts is a lower CO2 footprint, but in practice, it’s quite time-consuming (and nearly impossible) to construct perfect megaprompts that account for everything, and you will most likely still end up needing to elaborate and narrow down through continued dialogue. Often, the most fruitful approach is to use a more dialog-based method, where you continuously formulate new questions and understandings until you’ve covered everything.
Additionally, there are several more or less fixed methods you can base your prompts on. Again, it’s important not to take them too literally, but they can serve as useful guidelines.
One of the simplest prompt frameworks is the so-called R-T-F, where you give the chatbot a role, a task, and a format for the answer:
- Role (R), e.g., "You are an expert in project management"
- Task (T), e.g., "You need to create a project plan"
- Format (F), e.g., "The plan should be presented as a list"
Chain-of-thought (CoT)
It can also be a good idea to ask the chatbot to think through its response step-by-step or prompt it with a sequence of actions, called "chain-of-thought" prompting. This works especially well for tasks involving problem-solving, logic, or complex analysis, as it makes the model more aware of sequence and dependencies.
Remember that the technology is constantly changing, and prompt engineering is not an exact science, so try things out, experiment, be creative, and make your own experiences.
Prompts for studies
You don’t need to reinvent the wheel every time you write a prompt! Here you’ll find some inspirational examples that you can freely use and modify. The prompts are created using the R-T-F framework and chain-of-thought method. They are categorized according to the topics found here on study knowledge and can be used with your preferred AI study companion/language model.
Group work and planning
-
Reports and analyses
AI can serve as a secretary in group work. As a study group, when discussing topics such as assignment content, you can use AI to help structure and analyze your discussions based on pre-defined themes. However, if you do this, make sure not to upload data that contains personally identifiable information.
With the entire group’s consent, the meeting should be recorded, for example, through Word, which also offers direct transcription. The transcription is then inserted into AI with a prompt asking AI to analyze and summarize it based on the group’s pre-defined themes.
Example of a prompt:
After this prompt, I will insert a transcription from a meeting where we had [insert number] participants. After the full transcription is inserted, you should do the following:
- Read through the entire transcription.
- Then, identify statements from the transcription and categorize them into the following three themes:
- [insert theme 1]
- [insert theme 2]
- [insert theme 3]
- Afterward, organize the entire transcription according to the three themes listed above and provide a clear summary.
-
Create a study plan
You can use AI to help you draft a study plan. Prompt it with your fixed activities, such as classes, group meetings, etc., and tell it how many hours you will spend on reading and preparation in addition. You can also provide specific topic areas or literature you need to prepare and ask the generative AI to create a plan for you.
Example of a prompt:
You are a structured planner, and you need to help me create a study plan. The plan should take into account classes, group meetings, and leisure activities. Follow this process: Consider my fixed activities: [insert fixed activities with day and time]. The study plan should be created based on [insert lesson plan, reading list, or topic]. Based on the above, create a focused study plan.
-
Break down a project into tasks
A large project can seem overwhelming. You can use AI as a planning tool to break down a large and unmanageable project into smaller, manageable pieces.
Example of a prompt:
You are an experienced project manager, and you need to help me create a project plan for this project: [insert project description]. The plan should be action-oriented, realistic, and meet the deadline.
Assignment writing
-
Prompt for drafting a problem statement
You can use AI to draft a strong problem statement that includes the levels required for an assignment at a higher education institution. You can copy/paste the full prompt into your preferred AI. Note that Copilot is not suitable for this prompt.
Example of a prompt:
You need to help me formulate the problem statement for my final project at a business academy. The problem statement must include explanatory, analytical, and perspective levels. You will follow this process:
- Your first response should be to ask me what my project is about. I will provide an answer, and you will then help me improve it by following the next steps.
- Based on my answer, you will provide:
a) A revised and rewritten problem statement that is clear and easy to understand.
b) Suggestions on what I should add to the problem statement to improve it.
c) Relevant questions about what additional information I need to provide to make the problem statement better. - This process will continue, where I provide information and you ask questions, thereby improving and updating the problem statement until it is complete.
-
Prompt for structuring an academic assignment
When writing an academic assignment, it can be challenging to get an overview of the subtopics to include. You can use this prompt to generate inspiration based on your research question. However, always consult your advisor and study guidelines to ensure your assignment meets the necessary requirements.
Example of a prompt:
I am writing an academic assignment at a business academy. Provide me with suggestions on how to structure my assignment based on the following research question: [insert research question].
You should:
- Provide a list of subtopics to include in my assignment.
- Suggest what content should be covered within each subtopic.
Research and sources
-
Prompt for research support using AI
When you begin your research, it’s a good idea to work with your keywords. This can be done systematically by developing a mind map (read more about it here), and in this process, AI can be a creative tool.
AI can provide suggestions for keywords, supplement with other relevant themes based on a defined theory, and translate everything into English. The latter is a good idea since many databases are in English and therefore require searches in English.
Below we show different suggestions for prompting keywords for literature searches:
The first example is both context-based on your education and semester, but also theory-based, so AI is forced to respond based on a more theoretically well-founded basis.
Example prompt
You are a student at KEA, Copenhagen School of Design and Technology, studying [insert program]. You are in your [insert semester] semester, and in connection with a written exam assignment, you need to develop a mind map to find relevant literature. So far, you have the following keywords: [insert keywords separated by commas]. You now need to do the following:
- Find 2-3 synonyms for each keyword.
- Find 2-3 other relevant themes.
- All your suggestions in points 1 and 2 must be based on [insert your theoretical perspective].
-
Research from a problem statement
You can use this prompt to expand your problem statement, providing you with a range of relevant keywords for your continued search for literature.
The prompt can be copied and pasted into your preferred AI tool. Just remember to insert your problem statement.
Example prompt
You need to help me develop my search strategy for my final bachelor-level assignment at a business academy. You should base it on my problem statement: [insert problem statement].
- Identify key concepts from the problem statement.
- Then, find synonyms, related terms, and alternative spellings (e.g., British vs. American English).
-
Source criticism
You can use this prompt to evaluate a source. Insert the link to the source after the prompt. Always use your critical thinking when selecting sources.
Example prompt
Help me evaluate this source. It will be used for an academic assignment at a business academy. You should assess it based on its applicability regarding the sender, timeliness, argument, relevance, motive, and credibility. [insert source/URL]
-
References
You can use AI to create references for your bibliography from the sources you use in your assignment. This prompt will format the references according to the Harvard standard. If you want to use another standard, such as APA, just change it in the prompt.
You must provide the author name(s), publication year, book title, city/country of publication, publisher name, and optionally page numbers or book chapters.
Example prompt
You need to create a correct reference for a book following the Harvard standard:
Author name(s):
Publication year:
Book title:
Edition:
City or country of publication:
Publisher:
Optional page number or chapter:
Exam
-
Test your knowledge
You can use AI to test your academic knowledge in a specific subject. Consider uploading or linking to the course description or syllabus to provide context to the language model.
Remember, you can find all KEA course descriptions on KEA's catalog.
Example Prompt
You are a teacher and my personal tutor at a business academy, and you need to help me prepare for an oral exam. Please follow these steps:
- Ask for a description of the course. The description may be in a link, file, or other format and should include information about the exam format and required reading.
- Read the description thoroughly and ensure that the subsequent conversation stays within the boundaries of the course description.
- Ask me an academic question related to the course.
- Provide constructive feedback on my answer and explain how I can make my response more precise.
- Ask me a new question after giving feedback.
-
Find weaknesses in your report before an exam
If you need to defend a written assignment orally, you can use AI to prepare for the exam.
Be cautious about uploading your entire assignment to a chatbot and never upload anything containing personal data. You may consider uploading individual sections to give the chatbot more context to work with.
Example prompt
You are the examiner, and you need to help me prepare for an oral exam based on a written assignment. You should identify weaknesses in the assignment and help me prepare good responses as well as address key or missing parts that need to be covered during the oral exam. Here is the written assignment: [insert excerpt from the assignment].
-
Draft your presentation
AI can help you draft your presentation. This way, you're ready to go!
Example prompt
You are a clear, sharp, and effective communicator. You need to help me prepare for an oral defense of a written assignment and create a draft of the presentation. The presentation should be based on [insert assignment/topics/outline].
Questionnaires and interviews
-
Test your questionnaire
You can use AI to qualify your questionnaire before you send it out. Often, you would test with someone from the target group, but with AI, you can do some great preliminary work.
Example prompt
You are [insert target group] and need to help qualify a questionnaire. For each question, you should:
1. Explain how you, as a representative of the target group, understand the question and any potential misinterpretations that could arise.
2. Assess whether the response categories are balanced and exhaustive.
3. Suggest improvements or changes to enhance the questions.
[insert questions] -
Analyse your data
With AI, you can apply well-known analysis methods and prompt the AI to use specific methods. Examples of popular methods could include Pareto (80/20 rule), sentiment analysis, or coding interviews.
Note: Be mindful not to upload data containing personally identifiable information, and always consider whether you want to share data.
The prompt below helps to identify outlier responses.
Example prompt
You need to help me analyze data from a survey. Identify unusual or outlier responses that differ from the main trend in the answers and provide a summary. [insert relevant data]