19 September 2024 - Artificial Intelligence (AI)AI for All: URZ Establishes a Language Model Platform for the University
Since the launch of ChatGPT in November 2022, there have been rapid developments in the market for generative artificial intelligence applications and their performance continues to improve by leaps and bounds. This dynamic development has led to the emergence of a number of Large Language Models (LLMs) worldwide, including powerful open-source systems that are being utilized by large companies. On behalf of Heidelberg University, the URZ is now making this technology available to students and employees.
The platform “YoKI” now offers the opportunity to try out such an open-source language model in its initial test phase. In the future, the AI platform will offer a range of language models that will cater to the various needs of research, teaching and administration. This will enable users to choose the model that is best suited to their application.
Large Language Models (LLMs) and how they work
Large Language Models (LLMs) are a type of machine learning method and are categorized as generative AI tools. Interaction with a chatbot takes place in the form of a dialog or a question-and-answer session via prompts in the input field. Large volumes of text data can be quickly processed and new texts can be generated, translated or summarized. The created texts do not result from any thought processes but are rather the result of a probability-based compilation of the training data with which the program was fed in advance (pre-trained).
In practice, this means that such language models can provide false information – otherwise known as “hallucinations” – and are therefore not suitable for all areas of application. Facts should always be checked by the user. Even inside academia, the production of texts is no longer the sole preserve of human authors. Limitations such as AI hallucinations, toxic or discriminatory language, energy consumption and other ethical issues are just some of the risks that should be considered when using such programs.
AI Infrastructure on University Servers: Applications On Campus
The university AI infrastructure has been designed so that the systems are operated on premise, i.e., locally on our own servers on campus, which provides the following advantages:
- The prompts are not saved after the chat is closed and are not used for training and further developing the models.
- Personal data and information remains confidential because no external applications are used and third parties have no access to live chats.
- This service can only be used from within the University Network or via VPN.
