My research interests lie in developing and deploying AI technology for all language users in our multilingual society. To this end, I focus on inclusive LLMs and responsible AI, specifically:
building multilingual Large Language Models (LLMs) such as mT0, BLOOM and Aya,
supporting low-resource NLP by generating synthetic data (LexC-Gen) or parameter-efficient adaptation of pretrained LLMs (BLOOM+1), and
making LLMs safer and more representative for all language users. For instance, we discovered that low-resource languages can jailbreak GPT-4 and multilingual LLMs aren’t capable of code-switching.