Juliane Verwiebe
PhD candidate · Bethge Lab · University of Tübingen
About
I am a PhD candidate supervised by Matthias Bethge and Jakub Macina (ETH Zürich). I am a scholar of the International Max Planck Research School for Intelligent Systems (IMPRS-IS) and a member of the ELLIS PhD Program.
I am interested in uncovering how learning happens, how it can be measured at scale, and how LLMs can be used both to study it and to support it.
I hold an M.Sc. in Industrial Engineering and Management from Technische Universität Berlin, with a focus on machine learning and data science. Before my PhD I worked as a Machine Learning Engineer at Bosch Connected Industry in Berlin.
Research
LLMs are already used as tutors by millions of learners, yet there is little evidence that they actually help students learn. Current benchmarks mostly score individual tutor responses by proxies for good teaching rather than whether a whole interaction produces learning, and they largely rely on unrealistic synthetic dialogues.
My goal is to develop evaluation and training methods for LLM tutors, grounded in real human tutoring data and in measures of quality that take the whole interaction into account, validated against real learning outcomes. The central question is which features of an interaction predict genuine student progress, and how well scalable signals such as LLM judges or learned reward models track directly measured outcomes. The broader goal is to train LLMs that genuinely help students learn, not chatbots prompted to behave pedagogically.
- LLMs as tutors
- measuring learning outcomes
- RL post-training
Outreach
Beyond my research, I care about making AI more accessible and understandable. Through KI macht Schule, I teach workshops that help students understand the technology behind AI and develop the critical awareness needed to use it responsibly. I especially hope to encourage more girls to explore computer science and consider careers in the field. Previously I volunteered with Starcode e.V., giving coding classes for schoolgirls, and with the Berliner Lesepaten, reading with primary school children.
Selected publications
-
Personalizing Text-to-Image Generation to Individual Taste
Anne-Sofie Maerten*, Juliane Verwiebe*, Shyamgopal Karthik, Ameya Prabhu, Johan Wagemans, Matthias Bethge
P13N: Personalization in Generative AI Workshop, CVPR 2026 · arXiv
* equal contribution. Earlier work on stream processing is listed on Google Scholar.
Presentations & posters
-
Personalizing Text-to-Image Generation to Individual Taste 06/2026Poster · P13N Workshop, CVPR 2026, Denver