Woojung Song
M.S. in Data Science, Seoul National University
Research Interests
I work on AI agents and value alignment. As LLMs have become increasingly capable, they are entering everyday life through tool use, autonomous decision-making, and direct interaction with people. This raises a natural question: how do we ensure these systems behave in ways that reflect diverse human values, and how do we even know if they do?
- Tool-Use Agents. Enabling language models to effectively use tools so they can tackle more real-world tasks and assist users across diverse scenarios.
- Value Alignment. As LLMs take on more roles in society, we need ways to understand and evaluate what values they express. I work on building better frameworks for measuring this, moving beyond surface-level benchmarks toward methods that capture how models actually behave in context.
If any of this resonates, feel free to reach out at opusdeisong@snu.ac.kr.
Publications
Full list on the Publications page.
Proposes PA-Tool, a training-free method that renames tool schema components to align with small language models' pretraining familiarity, achieving up to 17% improvement on tool-use benchmarks without model retraining.
Shows that LLM psychological profiles from standard questionnaires diverge from profiles derived from actual generation behavior on real user queries.
A systematic framework to measure data contamination across item memorization, evaluation memorization, and target score matching in psychometric LLM benchmarks.
Proposes the Value Portrait framework for evaluating LLM values using items from real user-LLM interactions, finding across 44 LLMs that they emphasize Benevolence, Security, and Self-Direction.
