SDU is at the forefront of innovation and experimentation in the realm of generative AI, driving its adoption across education, research, and administrative functions.
Numerous initiatives are underway within faculties and departments to disseminate knowledge about this technology and empower staff to integrate it into their daily workflows.
New programs and resources are already being rolled out to benefit researchers, faculty, students, and other employees.
Experiences with the technology vary widely.
Some have quickly embraced it and integrated it seamlessly into their existing IT toolkit. Others have been slower to adopt it, either due to a lack of familiarity or an inability to see its direct relevance to their work.
Several colleagues have been surprised to find that the technology is not as straightforward as depicted in advertisements from tech companies promoting AI chatbots.
Crafting effective prompts requires a certain level of technical skill, and it takes time to become proficient with the tools.
This learning curve is not unique to generative AI and is similar to the experience of adopting any new IT system.
No, not yet. This is due to two primary factors: budgetary constraints and security assessments.
Currently, there are insufficient funds allocated to cover the additional licensing costs. Furthermore, Copilot 365 has not undergone the necessary security evaluations. These evaluations are essential to ensure the protection of documents and Teams, preventing unauthorized access to sensitive data.
No, not at this time.
Experiments are currently underway to develop SDU chatbots within AI Studio via Microsoft Azure, as part of the Generative AI in Administration project. These experiments are providing valuable insights into the requirements for creating effective chatbots.
Security assessments are also awaiting here – both for the system, but also for the chatbots that are being built.
If you wish to build your own chatbot for work purposes using services you have personally subscribed to, please adhere strictly to SDU's Guidelines for data and information security, as well as copyright regulations pertaining to generative AI.
Ask yourself these questions and be patient – this process takes longer than you might expect:
- What kind of chatbot do you want to build?
Define a clear purpose: Determine exactly what you want your chatbot to do. Should it answer frequently asked questions, provide customer service, or something else entirely? The more specific your purpose, the better you can tailor your chatbot.
- Which platform or tool will you use?
Choose a secure platform: Research how the platform/tool provider handles your data and select a secure option. A good rule of thumb is to use platforms/tools with data centers in Europe, as they comply with European data handling regulations.
- What kind of data do you have available?
Gather relevant data: Ensure you have a large and varied amount of data to train your chatbot. This could include conversations, articles, FAQs, or other relevant texts.
Quality over quantity: It is important that the data is of high quality and relevant to your purpose. Avoid inaccurate, outdated, or misleading information.
Vary the data: Try to include different styles, tones, and levels of complexity in your data to make your chatbot more robust.
Structure your data: Organize your data in a way that makes it easy for the model to understand. For example, you could use a format where each data unit consists of an input (e.g., a question) and an output (e.g., an answer).
- What features do you want your chatbot to have?
Experiment with parameters: Most language models have various parameters that you can adjust to optimize the chatbot's performance. Try different settings to find the best configuration.
Role-playing: Give the model a role or personality to create a more engaging conversation.
Safety measures: Implement safeguards to prevent your chatbot from generating harmful or discriminatory content.
Test and evaluate: Thoroughly test your chatbot with different types of input to identify potential issues and areas for improvement.
The SDU eScience Center is a dedicated unit committed to supporting research through the efficient utilization of data, modeling, and computation.
- The center provides a range of services and facilities to researchers, including the development of AI tools specifically designed for research purposes. The center is continually developing new AI tools and methodologies that can be employed to analyze large datasets, identify patterns, and inform decision-making. One such tool is U-Cloud.
U-Cloud is a platform facilitating interactive, high-performance computations, making it suitable for sensitive data analyses in full compliance with GDPR regulations. U-Cloud is built from the ground up adhering to rigorous security and data protection standards. All systems within the eScience Center undergo comprehensive security assessments conducted by the Center for Internet Security (CIS).
- SDU eScience has held ISO/IEC27001:2013 certification since February 2020, making it the first Danish university or public institution to achieve this certification.