Mădălina Zgreaban
Mădălina Zgreaban
PhD student
My PhD research focuses on building more complex and harder ways to test language models on their ability to understand language, such as inferring if two sentences entail each other, understanding texts, and answering questions. To build such tests, I leverage tools and resources from threebanks and grammars to large language models, working under the supervision of Lasha Abzianidze, Tejaswini Deoskar, and Yoad Winter. My other research interests are aligning or comparing language models to human values or rationales and explaining the inner workings, prompt engineering, and vision and language models.