AI Paper Atlas

Research map / Large language models / Alignment and preferences

Human-AI Curation Synergy: Scaling Preference Data Curation via Human-Guided AI Feedback

Chris Yuhao Liu, Liang Zeng, Yuzhen Xiao, Jujie He, Jiacai Liu, Chaojie Wang, Rui Yan, Wei Shen, Fuxiang Zhang, Jiacheng Xu, Yang Liu

ICLR 2026 · Large language models › Alignment and preferences · approach: Reward model design and preference models

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 simple LLM curation of unverified pairs (ablation)Skywork-Reward-V2 pipeline · Stage 1 iter 1, average of benchmarksaverage score: 0.1No clear effect
add human curation of unverified pairs (ablation)Skywork-Reward-V2 pipeline · Stage 1 iter 1, average of benchmarksaverage score: 2.3Improved
add adaptive retrieval of gold examples to LLM annotation (ablation)Skywork-Reward-V2 pipeline · Stage 1 iter 1, average of benchmarksaverage score: 0.9Improved
use human annotators with preference attributes instead of raw annotation (ablation)human curation · Stage 1 iter 1, average of benchmarksaverage score: 71.4 -> 72.1Improved
use full verification protocol with external tools instead of attributes only (ablation)human curation · Stage 1 iter 1, average of benchmarksaverage score: 72.1 -> 74.2Improved
raw human curation instead of no curation (ablation)human curation · Stage 1 iter 1, average of benchmarksaverage score: 71.0 -> 71.4Improved
scale uncurated preference data (ablation)SynPref-40M early version · average of six benchmarksaverage scoreNo clear effect
scale curated preference data (ablation)SynPref-40M early version · average of six benchmarksaverage scoreImproved
recycle discarded pairs by flipping chosen and rejected (ablation)Skywork-Reward-V2 · average of six benchmarksaverage scoreImproved
train on only 1.8% (about 290K samples) of the curated mixture (ablation)Skywork-Reward-V2-Llama-3.1-8B preliminary · average of six benchmarksaverage scoreImproved
add 14M flipped discarded pairs to the 26M curated pairs (40M variant)Skywork-Reward-V2-Llama-3.1-8B · seven benchmarks averageaverage score: 85.8 -> 88.6Improved
increase backbone from Qwen3-0.6B to Qwen3-8BSkywork-Reward-V2 · seven benchmarks averageaverage score: 70.9 -> 79.3Improved

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