Sofia de Trémiolles
I study the origins of political and social biases in language models
After completing a master’s degree in Data Science from CentraleSupélec and ESSEC, as well as completing a research thesis on social biases in multilingual large language models at Bocconi University, Sofia is now a PhD in NLP and social sciences, co-supervised by Djamé Seddah (INRIA, ALMAnaCH team) and Jean-Philippe Cointet (médialab, Sciences Po).
Her research focuses on political and social biases in language models. She studies the origins of these biases by analyzing the causal links between biases present in pre-training data and those that emerge when these models are used.