FAQ - KI:connect / RWTHgpt

• RWTHgpt users should contact the IT-ServiceDesk (servicedesk@itc.rwth-aachen.de) for any issues.
• KI:connect users should direct inquiries to the administrators of the respective participating universities. Since the system may behave differently depending on its configuration, the administrators are the first point of contact for questions or issues.

• GPT (Generative pre-trained transformer) describes an AI model that is able to create human-like texts and content (images, music and more) with the help of previously trained large language models.

• Tokens are the unit with which the AI models used measure the length of texts (both for user-generated questions and system-side answers).
  These are frequent character strings that occur in a text. The division into tokens is complicated and is not uniformly based on words or characters.
• Detailed information on calculating the number of tokens and the background can be found here:
  • OpenAI GPT Token Guide (opens in new tab) (DE)
  • What are Tokens (opens in new tab)
• OpenAI itself offers a calculator that makes it possible to determine the number of tokens and calculation in sentences or texts:
  • Tokenizer (opens in new tab)
    Please note that this page is subject to the general data protection guidelines of OpenAI. Please do not insert any personal data (from third parties) here.

• Yes, most RWTHgpt models have an interaction limit. However, models without interaction limits are also available. If necessary, you’ll need to switch to a model without a limit in the model selection.
   
• Administrators can set a custom interaction limit for each model. The limits may vary depending on the selected language model.
• The maximum number of interactions, as well as the number of interactions used so far, is displayed in the upper-left corner below the model selection field. A colored bar shows how much of the respective limits have already been used.
• In RWTHgpt, the limits for individual AI models are configured independently of one another. If a model’s limit is reached, you can simply switch to another model and continue working with it.
• RWTHgpt also offers models without interaction limits.

At present, it is not possible to adjust usage limits for individual users. The quotas are standardized for all employees or students and cannot be increased on an individual basis.

• Prompt is a term for a text input that is passed on to the AI.
• Both user input and system-side instructions can be described as a prompt.

• Detailed descriptions and notes can be found here:
  • Explanation of prompts and prompt engineering (opens in new tab) (DE)
  • OpenAI Prompt Engineering Guide (opens in new tab)

• Some of the language models available in RWTHgpt can recognize and analyze images. In the model overview (opens in new tab), you can check which models are capable of doing so.
• Image generation can be used in a new chat by clicking the Image button. Various image generation models with different interaction limits are available.
• Detailed information on image generation in RWTHgpt can be found in the article Image Generation.
• KI:connect supports image generation models. Usage permissions are configured by the administrators of the respective participating universities.

• Click on the paperclip icon to the left of the chat
• Select the image to be analyzed
• Now write a prompt that instructs RWTHgpt / KI:connect to analyze the image. You can also set restrictions here.
  • For example: “Analyze the attached image (focusing on the font and ignoring colors).