DISLab

We are working to bridge the alignment gap between LLM behavior and human expectations. Rather than refining optimization techniques, our lab prioritizes enhancing data quality to better incorporate human preferences into model generation. In particular, we focus on developing methods for curating high-quality datasets and utilizing them efficiently to achieve trustworthy and high-performance AI models. By advancing data-centric approaches, we aim to ensure that model outputs are not only accurate but also aligned with nuanced human values and expectations.