Talking about research in places where science is otherwise rarely present—that is the idea behind the event series “Heimspiel Wissenschaft.” For this, scientists return to their rural hometowns and present their work there. The Max Planck Institute for Informatics took part in the first Max Planck edition of these talks.
Olaf Dünkel, a doctoral researcher in the “Visual Computing and Artificial Intelligence” department led by Director Professor Christian Theobalt, visited his hometown of Boxberg-Schwabhausen in Baden-Württemberg on January 11, 2026, and gave a lecture on generative artificial intelligence titled “ChatGPT & Co: Generative AI for Images and Language.” Despite snowy weather, the topic was very well received: 86 participants came to the event at the village community center of the roughly 600-resident village.
Olaf Dünkel during his presentation. Photo: Albrecht Dünkel
For Olaf Dünkel, his “Heimspiel” was a very positive experience: “My aim was to educate people about generative AI and address widespread misconceptions, both overly positive and overly negative. It was quite challenging to prepare my work for a lay audience. But it worked well; the audience was very engaged, and after the talk there were many exciting follow-up questions and discussions.” The fact that the well-mixed participants included many familiar faces, from former kindergarden and school teachers to family and friends from his childhood, also made the evening something special for him.
The Max Planck edition of “Heimspiel Wissenschaft” is organized in cooperation with the Max Planck Society and the science communication agency Congressa. The series aims to bring researchers back to their rural hometowns to share their work and what it means to be a scientist. Speakers talk about what they’re working on, how and why they do their research, and how it connects to everyday life. The idea is to make science feel immediate and accessible, at the neighborhood pub, the community hall, the sports club clubhouse, or even the village square.
Currently, Olaf investigates how generative models can improve robustness in computer vision by leveraging learned priors and controllable data acquisition for training and evaluation. He also works on representation learning methods that better capture real-world correspondences. More broadly, he is interested in foundational questions that make intelligent agents reliable under distribution shift. Previously, he worked on safe ML, autonomous driving, and robotics.
Further Information:
https://heimspiel-wissenschaft.de/
https://people.mpi-inf.mpg.de/~oduenkel/
Editor:
Philipp Zapf-Schramm
Max Planck Institute for Informatics
Phone: +49 681 9325 4509
Email: pzs@mpi-inf.mpg.de