RWTHgpt features
The RWTHgpt service is based on the generative AI technologies of the company OpenAI and includes AI tools such as ChatGPT-3.5, ChatGPT-4o and the open source AI tool Mistral.
The systems can be accessed via a web interface developed at RWTH Aachen University, which enables data protection-compliant use of the OpenAI systems. The Login functions via RWTH Single Sign-On.
This article provides an overview of the functions available at the start of the RWTHgpt service:
- Introduction to RWTHgpt
- Terms of use and data protection agreements
- Selection of language models and AI tools *new language models are available since the latest update 26.08.2024
- Editing chats
- Opening a new chat
- Fold in navigation bar
- Editing prompts
- Upload documents
- Logout
RWTHgpt works similarly to other AI chatbots (and ChatGPT itself) as a text-based dialog system. As a user, you can ask the AI questions, assign text-based tasks or, depending on the AI tool or language model, have multimodal content (text & image) analyzed. AI content and answers may be incorrect, please always check important information. Instructions and tips on how to best pass on questions and tasks to the AI are provided here, for example:
- Prompts und Prompt Engineering (DE)
- Prompts und Eingaben bei OpenAI
- OpenAI Prompt Engineering Guide
- OpenAI Hilfecenter
2. Terms of use and data protection agreements
By using RWTHgpt, you agree to the general data protection agreements of RWTH Aachen University as well as the RWTHgpt terms of use (DE) and data protection agreements (DE). These can also be viewed in the RWTHgpt web interface by clicking on the question mark symbol on the right-hand side. Further information and a checklist for dealing with Generative AI systems (DE) can be found on the RWTHgpt project website.
3. Selection of language models and AI tools
In the RWTHgpt web interface, users can independently choose between different language models and AI tools. ChatGPT-3.5, ChatGPT-4o and the open source AI tool Mistral are available at the start of the service. Depending on the language model or AI tool selected, there are limits on the number of interactions or the total length of interactions. These can be seen in the selection and are highlighted in color while chatting. Further language models and AI tools will be made available after future updates of the RWTHgpt. An overview of the existing ChatGPT models and their functions can be found here.
On the left-hand side of the screen you will find an overview of all chats conducted with your registered account. By clicking on a chat, it can be reopened and used again. When a chat is resumed, the context of the previous messages in this chat is used again. When selecting a chat, it can be renamed and deleted. The default naming is carried out by RWTHgpt itself and refers to a system-side assessment of the first request made.
If this is desired, new chats can be opened using the plus button on the left-hand side of the interface. If a dialog has already taken place, it is then added to the list as another chat under "Heute"(today) and a new chat is opened without any entries. The context of previous messages is lost in the new chat. If the global token limit for a chat has already been reached, it is necessary to open a new chat.
If you do not need the navigation bar or want to enlarge the text area, you can fold it in by clicking on the small black line to the right of the navigation bar on the left-hand side of the screen. To expand it again, simply click on the black line again, which is now on the far left of the screen.
Individual prompts can also be edited. To do this, hold your mouse cursor over a message you have written and select the edit icon at the bottom right of the message. When a prompt is edited, the response from RWTHgpt is also regenerated and adapted. The context of the previous messages is taken into account. This can be used to rephrase or change the question or to continue using the same chat after the global token limit has already been reached.
Documents can be uploaded using the paperclip symbol to the left of the text input. Depending on the selected AI tool or language model, multimodal content (text & image) can also be used.
Once content has been uploaded, RWTHgpt can summarize and describe it and perform further tasks. When uploading any documents, especially copyrighted material or personal data (of third parties), please note the information on Terms of use and data protection agreements on this page.
If you want to explicitly log out of your account, you can do this at the bottom left of the screen. To do this, click on your initials or your name and select Logout. Alternatively, you can also close the tab or browser. When you access the RWTHgpt web interface again, you will then have to log in again using Single Sign-On.
