Woojin Kim

Woojin Kim

Dept. of Electrical & Computer Engineering, Seoul National University

I am a second-year MS–PhD student in the AIDAS Lab at Seoul National University, advised by Prof. Jaeyoung Do.

My research centers on diffusion language models and on aligning large language models to human values, with a particular interest in pluralistic alignment.

Diffusion LMs Omnimodal Models Human values in LLM Pluralistic alignment

01 Research

🌀

Diffusion Language Models

Non-autoregressive generation, sampling dynamics, and where diffusion can outdo next-token prediction.

⚖️

Alignment to Human Values

Steering LLMs toward what people actually value — beyond a single averaged preference.

🧭

Pluralistic Alignment

Representing and reconciling diverse, sometimes conflicting values within one model.

02 Surveys I Maintain

🌀

Awesome Diffusion LLM

A curated, community reading list on diffusion-based language models — methods, sampling, and applications.

★ GitHub ↗
🧭

Awesome LLM Values & Pluralistic Alignment

Papers and resources on aligning LLMs to human values, with a focus on pluralistic alignment.

★ GitHub ↗

03 News all news →

04 Selected Publications all publications →

Distributional value profiling overview
NeurIPS 2026 · Poster
Values Are Not Single Labels: Distributional Value Profiling Across Groups and Contexts
Woojin Kim, Jaeyoung Do.
NeurIPS, 2026 · Poster
VALUEFLOW overview
ICML 2026 · Oral
VALUEFLOW: Toward Pluralistic and Steerable Value-based Alignment in Large Language Models
Woojin Kim, Sieun Hyeon, Jusang Oh, Jaeyoung Do.
ICML, 2026 · Oral
Token Timestep Allocation illustration
NeurIPS 2025 · Poster
Don't Let It Fade: Preserving Edits in Diffusion Language Models via Token Timestep Allocation
Woojin Kim, Jaeyoung Do.
NeurIPS, 2025 · Poster