Soyeon Kim

Visiting Research Scholar

Soyeon Kim is a PhD candidate at Korea Advanced Institute of Science and Technology (KAIST). Her research centers on Responsible AI, with a focus on building safe and reliable generative models through data-centric approaches. She addresses core challenges in fairness, privacy, and hallucination to enable the trustworthy integration of AI systems into society. Her work includes a framework for automatically benchmarking LLM hallucination (NeurIPS’24, Spotlight) and a generative framework that jointly ensures privacy and fairness (ICLR’25). Her recent research targets the temporal reliability of LLM systems: evaluating and mitigating hallucinations in time-sensitive QA (ICLR’26) and improving temporal awareness in retrieval for RAG (EMNLP’26).