Large Language Models
Our research on large language models (LLMs) focuses on advancing the capabilities, reliability and applications of these powerful AI systems. We work on developing more trustworthy LLMs by improving their explainability, fairness and robustness. Our efforts include techniques to uncover LLM functioning, enhance fairness in in-context learning and mitigate issues like shortcut learning and biases. We also investigate the theoretical foundations of LLMs, analyzing their training dynamics and emerging capabilities like in-context learning. Our work extends to practical applications, exploring LLMs’ potential in mathematical reasoning, creative problem-solving and secure data analysis. From enhancing LLMs’ ability to solve complex math problems to developing frameworks for DataFrame question answering without data exposure, we aim to push the boundaries of what LLMs can achieve while ensuring their responsible and effective deployment in real-world scenarios.