AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

AdAEM: An Adaptively and Automated Extensible Evaluation Method of LLMs' Value Difference

Jing Yao, Shitong Duan, Xiaoyuan Yi, Dongkuan Xu, Peng Zhang, Tun Lu, Ning Gu, Zhicheng Dou, Xing Xie

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Uncertainty quantification for LLMs

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
use AdAEM generated questions instead of human-created general questionsvalue evaluation questions · human evaluation, 300 questionsreasonableness: 8.7%Improved
use AdAEM generated questions instead of human-created general questionsvalue evaluation questions · human evaluation, 300 questionsvalue differentiation: 52%Improved
use AdAEM Bench instead of ValueDCG for question setquestion diversity · benchmark statisticsSelf-BLEU: 13.93 -> 13.42Improved
use AdAEM Bench instead of ValueDCG for question setquestion novelty · benchmark statisticssemantic similarity: 0.36 -> 0.44Hurt
use AdAEM Bench instead of SVS for question setquestion diversity · benchmark statisticsSelf-BLEU: 52.68 -> 13.42Improved
prime o3-mini toward a target valueAdAEM measurement · value primingvalue score change: +31%Improved
prime o3-mini toward a target value, effect on opposing valuesAdAEM measurement · value primingvalue score change: -58%Hurt
prime o3-mini toward a target value, effect on same-group valuesAdAEM measurement · value primingvalue score change: +17%Improved
increase optimization budget BAdAEM · informativeness vs diversityinformativeness score and topic diversityMixed
use AdAEM questions instead of SVS and ValueDCG for LLM comparisonvalue evaluation · LLM value comparisondistinguishability of LLM valuesImproved
use GPT-4o instead of GPT-4o-Mini as value classifiervalue judgement · target domainaccuracy: 92.60 -> 92.92No clear effect
increase LLM size from 8B to 405BLlama models · AdAEM BenchTradition and Universalism preferenceImproved

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