Biography

What should an AI system do when the world has changed? This question has shaped my research and my sense of purpose as a computer scientist, since I began my research at Korea University. I studied how AI models, especially language models, handle time-sensitive knowledge, why they become unreliable when facts or regulations change after the training cutoff, and how we can make them respond more responsibly. In my previous work, I developed benchmarks for temporal knowledge and explored how temporal knowledge is represented inside models through mechanistic interpretability, leading me to propose novel alignment method. What distinguishes my research is that I do not treat evaluation, interpretability, and training regimes as separate problems; I aim to connect them into a single framework for building trustworthy AI. In my PhD, I hope to devise time-aware AI agents that can recognize when their knowledge is outdated, retrieve fresh evidence, revise their beliefs, and remain well-calibrated under uncertainty. My long-term goal is to help build AI systems that are not only more capable, but more reliable, transparent, and genuinely useful in improving people’s lives.