AI Paper Atlas

Research map / Large language models / Alignment and preferences

Disentangling Consensus and Value-Specific Representations for Controllable Pluralistic Value Alignment of LLMs

JianKui Zhou, Jing Yao, Xiaoyuan Yi, Peng Zhang, Ning Gu, Zhan Hu, Xing Xie, Tun Lu

ICML 2026 · Large language models › Alignment and preferences · approach: Steering and correcting LLM behavior

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add consensus anchor and subspace to the value subspace (ablation)DisAlign (Llama-3.2-3B) · MIC MMLUaccuracy (%): 46.5 -> 62.1Improved
add consensus anchor and subspace to the value subspace (ablation)DisAlign (Llama-3.2-3B) · MIC ARC-Caccuracy (%): 50.2 -> 77.2Improved
add value-specific subspace to the consensus component (ablation)DisAlign (Llama-3.2-3B) · MIC special parts D∆accuracy (%): 30.5 -> 53.5Improved
add value-specific subspace to the consensus component (ablation)DisAlign (Llama-3.2-3B) · MIC MMLUaccuracy (%): 62.8 -> 62.1Hurt
add consensus removal to diagonal Fisher decomposition (ablation)diag Fisher merging · MIC, B. Random D.accuracy (%): 45.8 -> 58.3Improved
add consensus removal to Euclidean decomposition (ablation)Euclid-I merging · MIC, B. Random D.accuracy (%): 40.8 -> 46.9Improved
use Fisher subspace decomposition instead of diagonal Fisher (ablation)DisAlign · MIC, B. Random D.accuracy (%): 58.3 -> 59.6Improved
use Fisher subspace decomposition instead of Euclidean (ablation)DisAlign · MIC, B. Random D.accuracy (%): 46.9 -> 59.6Improved
use DisAlign instead of the strongest prompting baselineLlama-3.2-3B · MIC B. Random D.accuracy (%): 54.3 -> 59.6Improved
use DisAlign instead of Prompt-Align on full-dimensional ValuePrismLlama-3.2-3B · ValuePrism C. Full D.accuracy (%): 55.8 -> 56.2No clear effect
use DisAlign instead of FWA/DARELlama-3.2-3B · MIC open-ended generationGPT-4o value alignment, random (1-5): 2.4 -> 4.2Improved
use DisAlign instead of FWA/DARELlama-3.2-3B · MIC open-ended generationresponse quality (1-5): 3.4 -> 3.7No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy