The current media coverage presents ChatGPT very ambivalently, from the savior to the end of the world caused by AI. In these portrayals, an explanation of the underlying technology and the resulting strengths and weaknesses is usually too brief. This is precisely where the training and awareness lecture by in-house specialists from the areas of IT and AI and the DITF data protection officer came in.
First, Dr. Heiko Matheis, who has been active as an AI trainer for SMEs for many years, explained machine learning in its various forms. In particular, he clarified that AI solutions are already being used in several research and consulting projects at DITF.
He vividly demonstrated that ChatGPT builds on the repeatedly used model of the pre-trained generative transformer (GPT) and extends it with a user interface, so that an almost human dialog (chat) is created. For this purpose, techniques from the AI fields of natural language processing and machine learning are combined.
Guido Grau then used concrete examples to demonstrate the potential of ChatGPT. The tool can be used in many areas to simplify and accelerate routine activities. Whether it is a matter of answering standard mails or creating the basic framework for an agenda, a presentation or a social media post, in all cases ChatGPT immediately provides the first concrete result texts on which the user can build.
However, it is essential to check generated texts for correctness. Supposed facts delivered by ChatGPT can quickly turn out to be false and invented. This is because the strength of KIModels to work with uncertainties and possibly compensate for them poses a problem here. This was illustrated by Guido Grau's request to ChatGPT for support in a literature search. The result was deceptively real looking literature, none of which really existed. The data protection officer Carsten Linti pointed out further risks and dangers in dealing with ChatGPT. He made it clear that many legal questions have not yet been answered. He urged not to enter any personal data at ChatGPT. Also, no information about company and business secrets should be included in the communication, because all data entered by the user would be stored on servers outside the EU, as the IT specialist Ralph Stoltz explained.
The great interest of the employees and their participation in the subsequent discussion round made it clear how important events of this kind are. With this presentation, the DITF seamlessly continued the events of a similar nature previously held at the partner institutions AFBW and innBW.
