PhD
Chair of Statistics and Data Science for Social Sciences and the Humanities at LMU
Institut für Statistik
Ludwigstr. 33
80539 München
Biosketch
Sarah holds a Bachelor’s degree in Economics and Political Science from the University of Heidelberg. During her studies, she gained consulting experience through internships at PricewaterhouseCoopers and the Boston Consulting Group. In 2022, Sarah obtained a Master’s degree in Social Data Science at the University of Oxford. She then joined the LMU Chair of Statistics and Data Science in Social Sciences and the Humanities as a PhD student. Her work focuses on human-centred artificial intelligence. Sarah has been part of relAI since October 2022 and is the student representative of the current cohort.
relAI Research
Understanding and Mitigating AI Risks in Socially Relevant Applications
Sarah’s research focuses on AI safety, using mechanistic interpretability methods to understand and mitigate risks in socially relevant applications. One key area of her work investigates jailbreaks in large language models—modifications to a prompt that bypass a model’s built-in safety mechanisms, causing it to generate responses it would otherwise refuse. Since there are multiple ways to jailbreak a model, one of Sarah’s projects examines whether these methods exploit similar internal mechanisms. Understanding this is crucial for developing more robust defenses against jailbreak attacks. Another strand of her research explores how large language models encode associations related to human preferences. By analyzing these internal representations, she aims to assess the reliability of using LLMs as proxies for human opinions in surveys, providing technical insights into their potential biases and limitations.