New Method Aligns Frozen LLM Judges with Human Judgment Distributions
A new arXiv paper introduces Human-Residual Steering, an alignment method that adapts frozen LLM judges to match the full distribution of human judgments, including individual variation, rather than only aggregated labels.
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-10-30 04:00 · arXiv cs.CL
Human-Residual Steering: Aligning Frozen LLM Judges with Human Judgment Distributions