Skip to main content

Large language models – getting past the hype

How can large language models be used to achieve the next leap in products and software development? Impressions of the Passau Data Science Summit 2024 at the University of Passau and statements from researchers and practitioners in the aftermovie.

Welcoming the participants of the Passau Data Summit 2024: the organisors Professor Michael Granitzer (University of Passau, from left) and Dr Andreas Böhm (One Data GmbH) together with Professor Harald Kosch, Vice President of the University of Passau. Copyright: Christian Franz | One Data

At the start of the Passau Data Science Summit, which focused on the future of data strategies, co-organiser Professor Michael Granitzer first looked back into the past. In the 1990s, former students managed to beat chess grandmaster Garry Kasparov with the help of a computer. The Deep Blue project made headlines around the world and demonstrated the potential of artificial intelligence for the first time.

Now, almost 30 years later, machines are not only beating humans at chess, but have also developed impressive language skills. The potential of large language models was a recurring theme at the Passau Data Science Summit, which brings together researchers and experts from leading international companies every year to discuss new trends and developments in the field of data strategies and products. The event is organised by the Chair of Data Science at the University of Passau together with the Passau-based company One Data GmbH.

Professor Harald Kosch, Vice President of the University of Passau, welcomed the more than 300 participants, including 160 representatives from industry. He described the importance of large language models as an issue that is becoming increasingly explosive. "Whoever controls the data controls the world."

AI models without much impact so far

Founder and Managing Director of One Data GmbH, Dr Andreas Böhm, himself an alumnus of the University of Passau, focused on the challenges for the economy in his keynote speech. After last year's hype, a certain disillusionment has set in. Contrary to predictions, AI-models have not yet led to a major leap in productivity. "How do we manage to unleash the transformative power of AI?" According to Böhm, the biggest problem is that the data is not yet being utilised properly. "The following applies to all AI models: the results are only as good as the underlying data. This year's PasDaS has impressively demonstrated that we can only realise the true value of AI when we combine it with high-quality company data and use it to create valuable data products," Böhm explained.

What the experts say

Playing the video will send your IP address to an external server.

Show video

How this can be achieved was one of the topics of the presentations and discussions at the conference. In the video, researchers from the University of Passau and practitioners talk about the potential and challenges of large language models:

  • Professor Steffen Herbold, holder of the Chair of AI Engineering, looks at the impact of generative language models on software development. He emphasises their disruptive effect: "In the past, there was the human programmer who had to write a source code himself, no matter how simple or complicated. Now, with language models, there is a new technology that is capable of solving simple tasks relatively reliably."
  • Ziye Wang, Head of Product Management at One Data, is one of the experts from the field who use generative language models in product development. She emphasises the advantages of this approach. The language models not only support her in her work when generating code, but she also uses them to structure data sets.
  • Professor Annette Hautli-Janisz holds the Junior Professorship of Computational Rhetoric and Natural Language Processing at the University of Passau. She researches what machines know about language - and what they don't. Complex contexts, for example, cause difficulties for machines.
  • Professor Jörg Schlötterer, who studied and completed his doctorate at the University of Passau, heads the hessian.AI junior research group at the University of Marburg and holds a deputy professorship at the University of Mannheim. With the help of his junior research group, he is trying to understand how knowledge is generated in machine learning models in order to develop language models that are tailored to different target groups.
  • Professor Florian Lemmerich, who holds the Chair of Applied Machine Learning at the University of Passau, is investigating how the quality of generative language models can be evaluated. This is difficult not only because the language models themselves provide very different answers to similar queries, but also because the development of the new technology is so dynamic.
Focus page

Large language models have disruptive effects. Researchers at the University of Passau are investigating the technical, social, ethical and legal consequences in an interdisciplinary manner.

About the Passau Data Science Summit

The Passau Data Science Summit (PasDas) has been held at the University of Passau since 2016. The two-day event is organised by the Passau Chair of Data Science in cooperation with One Data GmbH. The aim of the event is to bring together researchers in the fields of artificial intelligence and data science with leading experts from international companies. This year's guests included executives from Daimler Trucks AG, Deutsche Telekom, E.ON, OTTO GmbH, Schott AG, thyssenkrupp and the Volkswagen Group. This year's theme was "Deliver Business Value with Data Products". Professor Granitzer, who initiated the event in 2016 together with Dr Andreas Böhm, expressed his delight at its development in his welcoming address: it has now arrived in the largest lecture theatre at the University of Passau, the Audimax.

This text was machine-translated from German

Professor Michael Granitzer

conducts research in data science

How can contexts of meaning be identified in a deluge of digital media?

How can contexts of meaning be identified in a deluge of digital media?

Professor Michael Granitzer holds the Chair of Data Science. His research focuses on the use of machine-based learning methods and intelligent human-machine interfaces.

Playing the video will send your IP address to an external server.

Show video