University of Münster launches its own GPT platform "UniGPT"
From May 8, students and employees of the University of Münster can test the university's own chatbot UniGPT. The aim of this new AI-based service is to make various large language models (LLM) usable for studies, research, teaching and administration using modern language model technology.
As an innovative tool, UniGPT is designed to support university management, employees and students in making their daily work more effective and productive. By implementing the latest achievements in the field of language technology, the University of Münster is also promoting the progress of digital transformation in educational institutions. Fields of application for UniGPT can be found in various areas of everyday university life. Whether formulating emails, summarizing long texts, programming or translating – the service can be used for various tasks.
UniGPT is characterized in particular by model control and data protection compared to other GPT offerings. The service integrates both its own on-premise models, currently the open source model Mixtral 8x7B, as well as commercial models such as GPT-3.5 and GPT-4 from OpenAI. The University of Münster has control over the models used. As each model has specific strengths, users can select the best LLM for their application. Mixtral 8x7B is also operated in the Uni Cloud on servers at the University of Münster and is particularly suitable for sensitive data and projects where data protection is paramount. The integration of further models is planned.
During the test phase, which will last several months, reliability, availability and performance will be evaluated and improved under real conditions with many simultaneous users. Further functions such as web search and internal search are also expected to be released during this period. Comprehensive material on UniGPT, in particular instructions and background information, is currently still in preparation. Until then, we recommend that all interested parties join the official Mattermost channel for support, exchange of ideas and expert advice.
A university ID and the corresponding password are required to log in.