Information about the use of artificial intelligence
Our institution values data privacy and digital ethics. ChatUiT (https://chat.uit.no/) and Klartekst (https://klartekst.uit.no/) is approved for use with green, yellow, and red data. In line with the General Data Protection Regulation (GDPR), we have formulated comprehensive guidelines on how to use ChatUiT model in a manner that respects these principles. The following instructions relate specifically to personal information, sensitive information, copyright issues, and the use of ChatUiT in exams. We ask you to read them carefully.
On the 30th of November 2022, OpenAI launched a new large language model (LLM) ChatGPT (Generative Pre-training Transformer) version 3 for the public. Only three months later, a new, more advanced version of ChatGPT version 4 was released. Following a rapid development, and acknowledging risks rather than opportunities, the initial response of some universities was to ban the use of ChatGPT. However, in beginning of July, the UK’s Russel group created and signed five principles that will help universities to achieve “AI literacy” of their students and employees, acknowledging opportunities that LLMs such as ChatGPT can provide for teaching and learning. By introducing the ChatUiT service, UiT decided to follow these principles.
Launched on 2nd of October, ChatUiT is powered by GPT-3.5-Turbo and GPT-4. ChatUiT is approved for use with green, yellow, and red data expected in the very near future. The service is free for all of UiT’s students and employees and is accessed at chat.uit.no with UiT username and password.
But how does one “communicate” with ChatUiT? For this, we use prompts. A "prompt" refers to the input text or question that is provided to the model to initiate a response. It serves as a starting point or trigger for the model to generate output based on the information contained in the prompt. In simple terms, a prompt is an equivalent to a question that you would use when communicating with a human. For example, a typical “prompt” that students would use when communicating with a teacher is “Will this be in the exam?”
Similarly, you can ask ChatUiT to summarize a text for you, shorten it to X number of words, translate it to different languages. You can also ask it to explain some concepts or discuss them with you, to generate or debug a code in R or Python, and fan-favorite, to language edit a text. Both students and teachers can find out more on how to create prompts here.
Be aware that the model does not have the ability to fact-check information, as it is not a search engine like Google. It also does not have any inherent knowledge, nor does it possess the ability for original thought or deep contextual understanding. This means that when asked for some factual information (e.g. “What is the most cited paper in economy?”) it is very likely that it will give completely fabricated or biased information (also known as LLM hallucinations). In general, we would not recommend asking ChatUiT to provide accurate references for the statements it has made.
Both UiT students and employees have access to Bing Copilot (formerly known as Bing Chat Enterprise). Bing Copilot is an AI-powered feature within Microsoft's Bing search engine. It allows users to interact with Bing in a conversational manner, asking questions and receiving responses much like you would in a chat with a real person. This feature aims to make search more interactive and intuitive, providing not just links to websites but also direct answers, summaries, and even follow-up questions to help users find the information they need more efficiently. Bing Copilot is only approved for use with green data. Therefore, some guidelines apply (see below). You can find more information on how to access Bing Copilot here.
ChatUiT and Bing Copilot (for UiT) are fully GDPR compliant. Your prompts are not saved and the data you provide is not used to train the model further. However, as part of the design of Bing Copilot is that it will create and execute searches on the internet, some information could potentially leak through the search query.
For our students: Please use ChatUiT and Bing Copilot ethically and responsibly. Be aware of the guidelines on how to use it and the ChatUiT-related exam regulations (see below). We are confident that you will find the model helpful and that you will discover innovative ways to help you in teaching and learning. Always be aware of the model’s limitations! We are counting on you to provide your teachers with feedback on how you have been using ChatUiT, what was good about it and what was bad, what it was useful for and the instances when it was less useful. This feedback will be crucial to enhance teaching strategies, curriculum development, and responsible technology integration based on insights into its effectiveness and impact on student learning.
For our employees: Regardless if you are a teacher, administrative worker, or a researcher, please use ChatUiT and Bing Copilot ethically and responsibly. Be aware of the guidelines on how to use it and the ChatUiT-related exam regulations (see below). Always be aware of the model’s limitations! We encourage you to be open about it with your students and we encourage you to allow your students to use it in take home exams if possible. You can find a nice example of how you can do it here. We expect that you will find ChatUiT useful in various ways and contexts. If you have discovered innovative ways to use ChatUiT, please share them with your colleagues and with us. We would love to hear about them - send an email to: ki@uit.no! We hope the model will help you save time and effort, and that it will enhance and improve the quality of your teaching and research.
Finally, on this webpage, UiT AI team will publish regular updates on what is new when it comes to AI-related services offered by UiT, updates on guidelines, relevant resources, and planned AI-related activities.
Have fun with ChatUiT!
UiT AI team
Klartekst is a transcription service based on artificial intelligence that uses a language model called Whisper to carry out transcription. Whisper is an advanced, neural network-based, transcription model developed by OpenAI. It is designed to transcribe speech to text with high accuracy, even under challenging conditions such as noise or dialects. In the Klartekst service, users will be able to upload audio or video files where the Whisper model will analyze and convert the speech content in the files to written text. This service will be especially useful for transcribing interviews, where accuracy and the ability to handle varied speech are critical. Klartekst is approved for use with green, yellow, and red data.
Klartekst supports the National Library's NB-whisper model, which is trained on their content and is therefore better at Norwegian/Nynorsk speech and dialects. Read more about the National Library's work here.
Klartekst is approved for green, yellow and red data
Log in to the service at klartekst.uit.no and upload your files for transcription. For tips and help, see knowledge articles in the right-hand menu on this page.
Keenious is a tool designed to assist university students and staff in enhancing their research experience. Using AI algorithms, Keenious analyses text documents and PDFs, making it an excellent resource for finding relevant scholarly articles and research papers. It recommends literature based on the content of a user’s document rather than relying solely on keywords, providing a more personalized and in-depth research aid.
Available for both students and staff.
Download: Pre-installed in Word with a UiT account.
Personal information per now should not be entered in Bing Copilot
Personal data, as defined by GDPR, is any information that relates to an identified or identifiable living individual. Different pieces of information, which collected together can lead to the identification of a particular person, also constitute personal data.
To ensure compliance with GDPR, avoid sharing any identifiable information. This can include but is not limited to:
- Full name
- Home address
- Email address
- National identification number (Same as Social Security number in the U.S. or personal identification number in other countries)
- Passport number
- IP address
- Vehicle registration plate number
- Driver's license number
- Date of birth
- Birthplace
- Telephone number
- Credit card information
- Digital identity, like usernames or screen names
- Physical appearance or characteristic descriptions
- Genetic data
- Biometric data
Thus, to ensure data protection, do not share personal information such as those listed above, or any other data that can directly or indirectly identify you.
It's important to note that even in a group of data that may seem non-identifying on its own, if it can be combined to identify an individual, it's also considered personal data. For instance, sharing your job title might seem harmless, but if you're the "associate professor at the Department of community medicine, UiT” that could be used to identify you.
Keep in mind, data protection principles apply not only to the data that qualifies as personal but also to all the information that can be traced back to a person, be it by direct or indirect association.
GDPR defines sensitive information as data revealing racial or ethnic origin, political opinions, religious or philosophical beliefs, trade union membership, genetic data, biometric data for uniquely identifying a natural person, data concerning health, or data concerning a natural person's sex life or sexual orientation. Refrain from inputting this kind of sensitive information into Bing Copilot.
OK: "Can you explain the origins of Buddhism?"
NOT OK: "I am a practicing Buddhist. Can you explain the origins of Buddhism?"
While using ChatUiT, it is crucial to adhere to copyright laws. You should not input copyrighted materials into the model unless you own the copyright or have obtained necessary permissions.
OK: "Can you tell me about the main themes in George Orwell's '1984'?"
NOT OK: Inputting large sections or complete works of copyrighted materials, such as pasting an entire chapter from '1984'.
Prompting refers to the act of providing a text input or "prompt" to a large language model (LLM), such as ChatGPT, which then generates a response based on that input. The prompt acts as a starting point or instruction for the LLM, telling it what kind of information or output is being sought out. This could be a question, a statement, or a command. The quality and specificity of the prompt can greatly influence the relevance and accuracy of ChatGPT's response. Effective prompting is the key to getting the most out of conversational LLMs.
What is a good prompt?
A good prompt for generating text with ChatUiT should be clear, concise, and specific to the type of the response you want. Here are a few examples depending on different contexts:
Informational Prompt: "What are the main differences between renewable and non-renewable energy sources?"
Creative Writing Prompt: "Write a short story about a lost astronaut who discovers a new planet inhabited by a friendly alien civilization."
Instructional Prompt: "Explain how to perform a clean installation of the Windows 11 operating system."
Opinion-Based Prompt: "What are your thoughts on the impact of social media on face-to-face communication skills?"
Research Guidance: "Can you outline the steps for conducting a literature review in the field of psychology?"
Educational Prompt: "Provide a summary of the French Revolution, including its causes, key events, and outcomes."
Teaching Support: "What are some innovative methods to engage first-year students in large lecture classes?"
Student Advising: "What advice should we give students about managing the workload of a double major in business and computer science?"
University Administration: "Outline the steps needed to propose and implement a new campus-wide sustainability initiative."
Code Interpretation Prompts: "Explain what this Python function does and identify any potential errors in the logic."
Result Interpretation Prompts: "The regression analysis returned an R-squared value of 0.85. What does this indicate about the model's performance?"
Debugging prompts: "The output of the program includes several error messages after execution. Can you interpret these messages and suggest potential fixes?"
Language editing prompts: “Remove redundant words in this text, correct grammar mistakes, and make the sentences coherent.”
Prompts for translation: “Translate this text to French.”
Summary prompts: “Summarize this text in 200 words.”
What to think about when making a prompt?
The effectiveness of a prompt also depends on the depth of relevant information you provide. For example, if you are asking for advice, providing context will result in a more tailored and useful response. Similarly, if you're seeking creative content, giving some parameters or themes can help guide the ChatUiT in generating the desired output. In general, the more information you provide in the prompt, the better response you will get. You should think if you need to assign the role to the model, and specify if there is any specific behavior, tone or style that you want from the model in its response. For example:
Clarity and Specificity:
✅ Good Prompt: "Explain the process of photosynthesis in plants."
Context Setting:
✅ Good Prompt: "Given the recent advancements in renewable energy, can you explain the role of solar panels in reducing carbon emissions?"
Tone and Style:
✅ Good Prompt: "Write a casual, friendly response explaining the benefits of mindfulness meditation."
Length and Depth:
✅ Good Prompt: "Provide a detailed explanation of Albert Einstein's theory of relativity in a few paragraphs."
Open-Ended Questions:
✅ Good Prompt: "Discuss the impact of climate change on ecosystems
Multi-Part Prompts:
✅ Good Prompt: "In three parts: Explain the causes of World War I, the major events during the war, and its consequences."
Feedback and Iteration:
✅ Good Prompt: "Your previous response was informative, but can you also include some real-world examples to illustrate your points?"
Domain and Knowledge:
✅ Good Prompt: "In the context of computer science, explain the concept of 'algorithm efficiency.'"
It is very important to set your expectations. The initial response that you get from the model might not be as desired. It is important not to give up and adapt you important accordingly. Sometimes, several iterations of a prompt are needed in order to get the optimal response.
Finally, it is important to know that your chats are saved in the history panel to the left, and that you can always revisit them and pick up on your conversation with ChatUiT.
16.02.2024 UiT launches Klartekst for tale-til-tekst, automatic speech recognition service!
16.02.2024 Bing Copilot now available for students at UiT!
10.02.2024 AI-team arranges Webinar on ChatGPT in higher education on Monday 19.02.2024.
KA I KI? is a series of podcasts from UiT The Arctic University of Norway that explores the impact of artificial intelligence on education and research at Norwegian universities and colleges. The podcast features episodes discussing various aspects of AI, including:
- "Betyr KI slutten for samisk språk? Og for akademisk uavhengighet?" (Does AI mean the end for the Sami language? And for academic independence?) - aired on November 8, 2023.
- "Universitetets strategi for KI og studentenes perspektiv" (The university's strategy for AI and the students' perspective) - aired on October 3, 2023.
- "KI i forskning og i skoleverket" (AI in research and the school system) - also aired on October 3, 2023.
- "Undervisning og vurdering etter KI-revolusjonen" (Teaching and assessment after the AI revolution) - aired on September 2, 2023.
- "Dommedag for universitetene?" (Doomsday for universities?) - also aired on September 2, 2023.
The team behind the podcast consists of a historian, a linguist, and a philosopher, all affiliated with the Faculty of Humanities, Social Sciences, and Education at UiT Norges arktiske universitet.
Mission
Coordinate AI initiatives: The primary purpose of AI teams would be to coordinate all AI-related activities and initiatives within the university, ensuring they align with the institution's goals and objectives.
Develop and implement guidelines: It is crucial to establish clear and comprehensive guidelines for the use and development of AI. The team will formulate policies that promote ethical and responsible use of AI within the academic community.
Promote AI activities: The AI team will actively promote AI-related activities and encourage participation from students, faculty, and staff. This includes organizing workshops, seminars, and discussions to foster a deeper understanding of AI.
Enhance national collaboration: Strengthening ties with other universities and research institutions at the national level is essential. The AI team will collaborate with external entities to share knowledge, resources, and expertise in AI, and improve the overall academic and research environment.
Members
Marko Lukic, HelPed, Helsefak, leder av KI-teamet
Hedda Mørch, Institutt for samfunnsmedisin, Helsefak, nestleder av KI-teamet
Stig Brøndbo, Seksjon for forskning, utdanning og formidling Helsefak
Trygve Valnes, Eksamenstjenesten
Torstein Låg, Psykologi- og jusbiblioteket
Øystein Tveito, Seksjon for virksomhetsnære tjenester
John McNicol, Fakultet for humaniora, samfunnsvitenskap og lærerutdanning
Mariann Solberg, Fakultet for humaniora, samfunnsvitenskap og lærerutdanning
Lars Ailo Bongo, Institutt for informatikk, Fakultet for naturvitenskap og teknologi
Einar Holsbø, Institutt for informatikk, Fakultet for naturvitenskap og teknologi
Tor Wigum, Det juridiske fakultet
Kine Maridatter Dørum Maxwell, RESULT
Øyvind Hjuring Mikalsen, Seksjon for læringsmiljø og utdanningskvalitet
Pål Vegar Storeheier, Seksjon for forskning og innovasjon
Stig Brøndbo, Helsefak
Thorbjørn André Rydningen, Helsefak
Trygve Valnes, Seksjon for studieadministrasjon
Jonathan Crossen, Centre for Sami Studies
Jan Henriksen, UMAK
Hao Yu, Fakultet for ingeniørvitenskap og teknologi
Victor Zimmer, studentrepresentant
Contact: ki@uit.no
M365 Copilot is an advanced AI assistant integrated into Microsoft 365 applications to enhance productivity and efficiency. It helps users generate content, automate tasks, and gain insights from data. For example, M365 Copilot can assist with writing emails, creating documents, making meeting notes, analyzing data, and much more.
However, there are many prerequisites before using such a tool. Copilot has access to all the information you have access to in M365, including sensitive data, emails, and chat messages. Your Copilot does not have access to information you do not have access to.
UiT is currently conducting technical testing of M365 Copilot, and the test group consists of 15 people, including technical personnel, lawyers, and the data protection officer. The testing will continue until we have an overview and control of what happens with the data.
Technical testing helps identify and address potential security issues and ensure that your data is protected. This includes logging, opportunities, risks, various uses of prompting, and how it handles your data. Testing can also reveal if there are failures in access control.
There will be updated information on the way forward after the technical testing of M365 Copilot. If it is decided to proceed with the process, the next phase will be to conduct a limited pilot, focusing on functionality and utility.
NTNU and The Norwegian Data Protection Authority (Datatilsynet) have jointly conducted a project and tested Copilot for Microsoft 365 this spring. Information about the project, their findings, and the report can be found here if you are interested in more information: Copilot for Microsoft 365 in the public sector. (Only in Norwegian)