VALUEFLOW has been accepted to ICML 2026 as a Spotlight!

We are excited to share that our paper, VALUEFLOW: Toward Pluralistic and Steerable Value-based Alignment in Large Language Models, has been accepted to The International Conference on Machine Learning (ICML 2026) as a Spotlight.

VALUEFLOW introduces a unified framework for pluralistic value-based alignment, covering value extraction, evaluation, and steering with calibrated intensity control. It combines HiVES, a hierarchical value embedding space, the Value Intensity DataBase (VIDB), and an anchor-based evaluator to study and control how LLMs express values across multiple value theories.

Paper:

• VALUEFLOW: Toward Pluralistic and Steerable Value-based Alignment in Large Language Models - Woojin Kim*, Sieun Hyeon, Jusang Oh, Jaeyoung Do^

Congratulations to all the authors!