AI Paper Atlas

Research map / Large language models / Alignment and preferences

When Weak LLMs Speak with Confidence, Preference Alignment Gets Stronger

Amirabbas Afzali, Myeongho Jeon, Maria Brbic

ICLR 2026 · Large language models › Alignment and preferences · approach: Robust and calibrated DPO

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
reweight DPO samples by weak-LLM confidence (CW-DPO) instead of WS-DPO (ablation)DPO with weak labels · avg of HH-RLHF/TL;DR/UFB, OPT-13BGRA (%): 57.9 -> 60.3Improved
reweight IPO samples by confidence (CW-IPO) (ablation)IPO with weak labels · avg of 3 datasets, OPT-13BGRA (%): 57.9 -> 61.5Improved
reweight rDPO samples by confidence (CW-rDPO) (ablation)rDPO with weak labels · avg of 3 datasets, OPT-13BGRA (%): 55.5 -> 62.7Improved
confidence weighting on Qwen2.5-14B strong model (ablation)DPO/IPO/rDPO · avg of 3 datasets, Qwen2.5-14BGRA (%): 74.8 -> 75.6 (DPO), 73.7 -> 77.2 (IPO), 74.6 -> 77.3 (rDPO)Improved
confidence weighting on DPO, HH-RLHF with Qwen 14B (ablation)CW-DPO · HH-RLHF, Qwen2.5-14BGRA (%): 81.4 -> 80.6Hurt
use CW-DPO with 30% human labels instead of DPO on 100% human labelsCW-DPO · HH-RLHF, OPT-1.3BGRA (%): 71.9 -> 69.9Hurt
use CW-DPO with 30% human labels instead of DPO on 100% human labelsCW-DPO · avg of 4 datasets, Qwen2.5-7BGRA (%): 65.1 -> 70.6Improved
use CW-DPO with 20% annotations vs DPO on 100% human labelsCW-DPO · HH-RLHF, OPT-1.3BGRA (%): 69.7 -> 70.3Improved
use larger strong modelsCW-DPO · HH-RLHFGRA gain over baselinesMixed
use C1 weighting instead of sigmoid or clipped-margin weights (ablation)CW-DPO · avg of Harmless/Helpful/HH-RLHF, QwenGRA (%): 69.4 -> 72.7 vs C2; 68.4 vs C3; 70.1 vs C4Improved
train weak annotator with Bradley-Terry scalar head instead of DPO or SFT+DPO (ablation)weak annotator · unlabeled HH datareward accuracy (%): 53.8 (DPO), 54.9 (SFT+DPO) -> 64.8 (BT)Improved
use confidence weighting instead of top-N% confidence filtering (ablation)CW-DPO · HH-RLHFGRAImproved

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