AI Paper Atlas

Research map / Large language models / Alignment and preferences

Probing Preference Representations: A Multi-Dimensional Evaluation and Analysis Method for Reward Models

Chenglong Wang, Yifu Huo, Yang Gan, Yongyu Mu, Qiaozhi He, Murun Yang, Bei Li, Chunliang Zhang, Tongran Liu, Anxiang Ma, Zhengtao Yu, Jingbo Zhu and 1 more

AAAI 2026 · Large language models › Alignment and preferences · approach: Reward model design and preference models · 3 citations (OpenAlex)

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 inference-time-probing sample filtering to PPOLLaMA-3.1-8B-RM-400k PPO · Aligned LLaMA-3.1-8B-SFTwin rate: 57.3 -> 62.5Improved
discard random samples instead of probing-selected onesPPO · Aligned LLaMA-3.1-8B-SFTwin rate: 57.3 -> 54.3Hurt
lower threshold d_tau from 140 to 100 (ablation)inference-time probing · Threshold sweepwin rate: 63.5 -> 59.6Hurt
raise threshold d_tau from 140 to 180 (ablation)inference-time probing · Threshold sweepwin rate: 63.5 -> 60.1Hurt
increase cluster size from 2 to 5 (ablation)inference-time probing · Cluster size sweepwin rate: 63.5 -> 59.2Hurt
use inference-time probing with any threshold vs vanilla (ablation)PPO · Threshold sweepwin rate: 57.4 -> 59.6Improved
fuse MRMBench with RM-Benchreward model evaluation · Correlation with aligned LLM win ratePearson correlation: 0.89 -> 0.92Improved
compare MRMBench with RewardBench as predictorreward model evaluation · Correlation with aligned LLM win ratePearson correlation: 0.34 -> 0.89Improved
use top-most layer instead of lower layers for probing (ablation)GPM-Llama-3.1-8B and GPM-Gemma-2B · MRMBench probing tasksaccuracyImproved
compare easy vs hard probing tasksGPM-LLaMA-3.1-8B · MRMBenchaverage accuracy: 79.6 -> 64.5Hurt

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