AI Paper Atlas

Research map / Large language models / Alignment and preferences

Verification and Co-Alignment via Heterogeneous Consistency for Preference-Aligned LLM Annotations

Cheng Chen, Haiyan Yin, Ivor Tsang

ICLR 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
apply HCC co-alignmentMeta-Llama-3-8B-Instruct · Clincaccuracy (%): 32.49 -> 82.43Improved
apply HCC co-alignmentMeta-Llama-3-8B-Instruct · Massive Scenarioaccuracy (%): 43.52 -> 78.13Improved
apply HCC co-alignmentGPT-4o Mini · Banking77accuracy (%): 65.12 -> 82.45Improved
apply HCC co-alignmentGPT-4o Mini · Clincaccuracy (%): 81.44 -> 87.93Improved
apply HCC co-alignmentGPT-4o Mini · MTOP Intentaccuracy (%): 75.03 -> 67.10Hurt
apply HCC co-alignmentGPT-4o Mini · StackExchangeaccuracy (%): 51.90 -> 45.22Hurt
add the correction (co-alignment) step on top of HCC without correction (ablation)HCC with GPT-3.5 Turbo · MTOP Intentaccuracy (%): 79.57 -> 69.06Hurt
add the correction (co-alignment) step on top of HCC without correction (ablation)HCC with GPT-4o Mini · MTOP Intentaccuracy (%): 80.16 -> 67.10Hurt
add the correction (co-alignment) step on top of HCC without correction (ablation)HCC with GPT-3.5 Turbo · Clincaccuracy (%): 81.32 -> 85.49Improved
use stronger MiniLM instead of BERT-Base as the specialised model (ablation)HCC with Llama-3-8B Instruct · Banking77accuracy (%): 71.62 -> 77.71Improved
use stronger MiniLM instead of BERT-Base as the specialised model (ablation)HCC with Llama-3-8B Instruct · Massive Scenarioaccuracy (%): 74.61 -> 78.13Improved
use imbalanced user-preference samples (60% imbalance) instead of balanced (ablation)HCC with Llama-3-8B Instruct · Clincaccuracy drop: -0.32Hurt

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