Bielefeld IT Service Centre
BIKI - Bielefeld AI Portal
An application for chats with large language models
The AI portal "BIKI" was developed to provide users of Bielefeld University with a provider-independent solution for using artificial intelligence based on a large language model.
A Large Language Model (LLM) is an AI model that is trained on large text datasets to understand and generate human-like language. It is based on deep neural networks and can perform tasks such as text generation, translation, answering questions and much more. LLMs use complex pattern recognition to provide contextualised and coherent responses.
Currently, ChatGPT from OpenAI and the upcoming language models from the Gesellschaft für wissenschaftliche Datenverarbeitung Göttingen (GWDG) are offered in BIKI:
Llama 3.1 (8B),
Llama 3.1 (70B),
Qwen2.5 (72B),
Mistral Large,
Codestral 22B.
The data entered in the GWDG language models are processed exclusively on servers in Germany.
The language models originate from various manufacturers, but even if a language model was originally developed by Meta (Facebook), for example, there is no data transfer to these manufacturers when used in BIKI via the GWDG.
The service is to be further expanded in the future.
The most important aspects of Data protection and information security:
- The solution is only suitable for input / data with normal protection requirements.
- During use, only the communication data is pseudonymised so that no conclusions can be drawn about the staff, people using the system. All content data, i.e. the user's entries, are transferred 1:1 to the OpenAI servers and stored there for 30 days.
- The LLM is not trained using the content data entered.
Scope of services
Risks and opportunities of artificial intelligence (AI)
Technical protective measures by BIKI
The most important technical measures to protect the information are
- Pseudonymisation (Data protection)
The BIKI web interface enables pseudonymisation of communication data (e.g. user name, IP address). BIKI does not allow LLM providers to assign content data based on communication or authentication.
- Persistence of chat histories (information security)
A key feature of LLMs is the ability to have different, parallel conversations and to access and continue them at a later point in time. Accessing the history of an individual chat is necessary if only to provide the language model with a chat memory and thus a 'short-term memory' on the basis of which it can give its answers with reference to earlier parts of the conversation.
BIKI enables this feature by storing the communication history locally despite pseudonymisation.
- Earmarking (Data protection)
The University contracts with external AI providers to ensure that the data you provide is not used for training the manufacturer's LLM.
Note on risks and protective measures when using Large Language Models (LLMs)
The use of Large Language Models (LLMs) offers many advantages, but also harbours risks that must be taken into account in accordance with the AI Regulation (EU AI Act):
- Impermissible risks: Certain applications of LLMs, such as social rating systems or systems that manipulate people, are prohibited. Always use LLMs responsibly and in accordance with ethical standards.
- Transparency requirements: If LLMs are used in applications that interact with humans (e.g. chatbots), users must be informed that they are communicating with an AI.
Protective measures:
- Data quality and security: In principle, no confidential, personal or sensitive official data may be processed in AI systems.
The use of AI must be in accordance with the Guidelines and Procedures for Safeguarding Good Scientific Practice at Bielefeld University dated 9 February 2023 and the EU Artificial Intelligence Act.
The university management of Bielefeld University is currently working intensively on the use of AI in business processes (research, studies and teaching, administration) and is developing corresponding service agreements together with the Faculties and staff councils.
Requirements and status
Target group
- Students from 1 October: Trial operation
- All teaching staff, instructors, lecturers: Trial operation
- Administrative staff: Selective trial operation
Background to the trial operation
It makes sense for many departments within a university to consider the use of artificial intelligence (AI) in their business processes, not just the IT department, as each department has unique requirements and challenges that require specific AI solutions.
A holistic approach enables the development of customised solutions that meet the specific needs of each department and improve the efficiency and quality of results.
In addition, the integration of all disciplines in the process of AI integration promotes interdisciplinary exchange, which leads to innovative approaches and new research ideas.
This ensures that the university's AI strategy is sustainable, future-orientated and aligned with the specific ethical, legal and social aspects of the various disciplines.
If you are an administrative employee and would like to take part in the trial run and agree to the terms of use, please contact the BITS Service Desk. Please describe how many staff, people will be using BIKI and for what purpose.
Background of the test operation
It makes sense for many departments within a university to consider the use of artificial intelligence (AI) in their business processes, not just the IT department, as each department has unique requirements and challenges that require specific AI solutions.
A holistic approach enables the development of customised solutions that meet the specific needs of each department and improve the efficiency and quality of results.
In addition, the integration of all disciplines in the process of AI integration promotes interdisciplinary exchange, which leads to innovative approaches and new research ideas.
This ensures that the university's AI strategy is sustainable, future-orientated and aligned with the specific ethical, legal and social aspects of the various disciplines.
If you would like to participate in the test operation and agree to the terms of use, please contact the BITS Service Desk. Please describe how many staff, people will use BIKI and for what purpose.
Data protection
- Data protection information (in German)
Terms of use
Please note that the use of BIKI is subject to the guidelines of the BITS rules of use. It is important that all users read these guidelines carefully and comply with them to ensure proper and responsible use.
Support
Service Desk
Mo.–Fr. 08:00–16:00 h
Tel.: 0521 106-6000
E-mail: servicedesk@uni-bielefeld.de
