HAWKI.HSZG

HAWKI was developed at the University of Applied Sciences and Arts Hildesheim/Holzminden/Göttingen (HAWK) and enables GDPR-compliant access to various AI chat models (including ChatGPT). Access at the HSZG is provided by the university computer center.

HAWKI attaches particular importance to data protection. The prompts (inputs) to the HAWKI interface are transmitted to the respective Large Language Models (LLMs) via an API (Application Programming Interface). This means that individual users cannot be assigned and no personal profiles can be created (neither user account nor IP address). The chat content is not used as training data for the Large Language Model (LLM).

Step 1: Call up HAWKI.HSZG and click on "Register"

Step 2: Register with university login

Step 3: Click on "Will do!"

Step 4: Now enter your own datakey

Step 5: Enter the datakey again and then confirm

Step 6: Save the recovery code

Step 7: Start with HAWKI.HSZG

Note: The datakey is set up only once. After that, you only need to log in with your university login.

Use is permitted for the following purposes:

• Programming and software assistance: assistance with writing, analyzing and understanding program code

• Language and text optimization: help with the linguistic improvement of texts (e.g. grammar, style, spelling)

• Idea generation and creativity support: brainstorming, developing ideas for projects, concepts or scientific questions

• Administration and organization: support with creating schedules, to-do lists or project management tasks

• Advice on ethical and data protection issues: support in raising awareness and reflecting on the responsible use of AI

Of the models that can be used by OpenAI, the "GPT-4o model" is a recommended language model. It is very powerful and can be used without restrictions. However, no personal or copyright-protected data should be entered as a prompt, as the processing of the input takes place on servers in the USA and could possibly be interpreted as duplication. You can also use"Llama 3.1 70B instruct" and "Alibaba Cloud Qwen 2.5", two "open source" language models hosted on servers in Germany (GWDG). However, the context window (scope of the prompts) is smaller with these models and content errors occur more frequently.

Alternatively, various open source language models are available via the portal of the Gesellschaft für wissenschaftliche Datenverarbeitung mbH Göttingen (GWDG-LLM-Service) (Academic Cloud). The"Meta LLaMA 3.3 70B Instruct" model and the "DeepSeek R1" model for tasks with logical-mathematical relationships should be highlighted here.

Access to HAWKI is only possible after completing the self-study course available in OPAL "AI Act: Introduction and sensitization for university staff" possible. Please note the information on compulsory training at the HSZG contained in Circular 2025/05 (07.04.2025).

Note: To obtain the certificate, registration with the HSZG login in OPAL is required. The integrated tests must be completed in full.