For around a year now, students and staff at Hof University of Applied Sciences have had access to their own AI tools for text and image generation (“Lisa and Luis”) , which are becoming increasingly popular. They were specially developed and made available by the AI User Center at the Institute for Information Systems (iisys). We spoke to Prof. Dr. Sven Rill about the current status.
How did the idea of developing your own language and image generation programs for Hof University of Applied Sciences come about and what challenges did you face?
“Now that ChatGPT & Co has been rapidly integrated into the everyday lives of students and the way it is used in certain areas is completely different to the way it was previously used with Google, it quickly became clear that we had to adapt to this and rethink and adapt. Together with Prof. Dr. René Peinl and our Competence Center for Applied Artificial Intelligence(KI-AWZ), we have resources that we wanted to use.”
Do others also rely on their own solutions?
“Other universities often rely on ready-made solutions from providers such as OpenAI with ChatGPT. We take a critical view of this, as we will lose some independence and, above all, expertise if we can’t “master” such technologies ourselves. If the AI tools are developed in-house, it is possible to offer specific knowledge and special functions for students in a targeted manner. This would be an “agent” specialized in a specific lecture, for example.”
After around a year of operation, what significant experiences have been made with it and what development steps can be reported?
“We started a year ago. At that time, we provided the systems for a chatbot and the image generation tool for students and lecturers at very short notice. User-friendliness, data protection and quality were important to us. Rapid progress is still being made in the field of large language models, which we naturally want to take advantage of. We are therefore already preparing the third version (LISA v3), which is based on a significantly improved language model compared to LISA v1. We have also made significant progress in terms of user-friendliness. LISA v3 is based on a new front end, which offers many new functions that we now want to make available step by step.”
What specific benefits do “Luis” and “Lisa” offer students, for example when preparing for exams or creating presentations?
“LISA is designed to help students prepare for exams. With the help of their own agents (customized chatbots based on the standard LISA v3 model), a mentor is to be created who, like the professor, can answer questions about the lecture material or generate exam questions. This should enable students to prepare for the exam.
LUIS makes it possible to generate a wide variety of images. This is very helpful for presentations in particular.”
Can you give further practical examples of how students can use “Luis” and “Lisa” in their everyday studies?
“LUIS is very good for generating images. Our media informatics students really enjoy using it. LUIS can also be used well in the field of marketing.
LISA can generate, summarize, correct, translate, etc. texts, but also help with the development of a new topic. Through targeted “inquiries”, new content and ideas can be gained. In the field of computer science, we use code generation tools. These are no substitute for a software developer, but they are very helpful for generating small pieces of code.”
What else distinguishes “Luis” and “Lisa” from established AI tools such as ChatGPT or DALL-E, which also enable speech and image generation?
“In terms of the quality of the results, established AI tools are often even better. They are designed to provide a pretty good answer to everything. With our own AI tools, however, we can address specific issues. However, if we integrate our own content (such as course content from Moodle, lecture notes, university data) into the AI tools, the results will be better. And as already mentioned, we see it as essential that we build up expertise in this area. If we don’t do this, we will lose the “connection” in the next few years and will only be able to use what is then offered by a few monopolists.”
One topic that interests students and the administration alike is data protection. How do you ensure that the programs are secure and GDPR-compliant?
“The topic of data protection plays another important role. The systems are all operated in the KI AWZ. This means that no data is processed outside Hof University of Applied Sciences. The data is neither reused nor analyzed in any other way. In addition, we regularly consult with the university’s data protection officer. This is different with ChatGPT & Co. Especially in the free versions, you often give consent “that you don’t know about”.”
Are there plans to integrate “Luis” and “Lisa” more strongly into everyday university life, for example through cooperation with the administration or through special training for students?
“We want to use the upcoming exam period to draw more attention to their use in preparing for exams. We want to use examples and templates to provide an easy introduction. We want to approach students with various information over the next few weeks.”
How do you see the further development of “Luis” and “Lisa” over the next few years? What new functions or areas of application can you imagine?
“We are currently discussing integration into our Moodle learning platform. Significant progress can be made if we make “our” data accessible through the AI tools. There will be a lot of technical developments. We will actively stay on the ball here and adopt innovations where possible.”
