AI Paper Atlas

Research map / Large language models / Alignment and preferences

GEM: Generative Entropy-Guided Preference Modeling for Few-Shot Alignment of LLMs

Yiyang Zhao, Huiyu Bai, Xuejiao Zhao

AAAI 2026 · Large language models › Alignment and preferences · approach: Robust and calibrated DPO · 0 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 Cognitive Filtering and SEGA (ablation)GEM on Llama-3-8B-Instruct · UltraFeedbackpreference-prediction accuracy (%): 69.0 -> 77.1Improved
add Cognitive Filtering and SEGA (ablation)GEM on Llama-3-8B-Instruct · GSM8Kaccuracy (%): 48.3 -> 55.6Improved
add final-answer entropy term to the token score (ablation)GEM · GSM8Kaccuracy (%): 50.1 -> 55.6Improved
add fork-token entropy term to the token score (ablation)GEM · GSM8Kaccuracy (%): 52.7 -> 55.6Improved
use SEGA instead of pairwise DPO on filtered CoTs (ablation)GEM · Medical QA expert agreementagreement (%): 73.0 -> 78.2Improved
use SEGA instead of pairwise DPO on filtered CoTs (ablation)GEM · UltraFeedbackpreference-prediction accuracy (%): 74.5 -> 77.1Improved
use GEM instead of DPO on few-shot preference pairsLlama-3-8B-Instruct · GSM8Kaccuracy (%): 50.2 -> 55.6Improved
use GEM instead of DPO on few-shot preference pairsLlama-3-8B-Instruct · RewardBenchpreference-prediction accuracy (%): 63.2 -> 75.4Improved
use GEM instead of Reward Model + PPOLlama-3-8B-Instruct · MT-Bench (win-rate vs SFT baseline)win-rate (%): 47 -> 68Improved
use GEM instead of Reward Model + PPO on medical QALlama-3-8B-Instruct · Medical QA expert agreementagreement (%): 72.5 -> 78.2Improved
add Cognitive Filtering and SEGA (ablation)GEM on Llama-3-8B-Instruct · Medical QA expert agreementagreement (%): 70.5 -> 78.2Improved

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