Jan Batzner

Jan Batzner is a Research Associate at Weizenbaum Institute Berlin, the German Internet Institute, and a Doctoral Researcher in Computer Science at Technical University Munich (TUM). His research focuses on Large Language Model Alignment. Jan graduated with a Master‘s in QMSS from Columbia University in New York City.

Before joining the Weizenbaum Institute, Jan worked for a leading hybrid cloud computing enterprise and has been recognized as a Fellow of the International Cooperation of Assigned Names and Numbers (ICANN) and the Bavarian Elite Academy (BEA). 

Current Project:

Sycophancy, deceptive alignment, and (harmful) short-cuts in the learning process in AI training, all relate back to the need of better measurement and benchmarks. Jan is working on developing LLM Alignment Benchmarks in order to address some of the harms that sycophancy and deceptive alignment can produce. Motivating talented junior researchers for building Sycophancy Benchmarks and being part of the AI Alignment community, benefits society overall. Ideally we do contribute to mitigation and advance to the current state of evaluation, but most importantly is to get researchers with this QMSS innovation lab involved and part of the process of making conversational AI safer.