RLHF-aligned language models exhibit response homogenization: on TruthfulQA (n=790), 40-79% of questions produce a single semantic cluster across 10 i.i.d. samples. On affected questions, sampling-based uncertainty methods have zero discriminative power (AUROC=0.500), while free token entropy retains signal (0.603). This alignment tax is task-dependent: on GSM8K (n=500), token entropy achieves 0.724 (Cohen's d=0.81).
A base-vs-instruct ablation confirms the causal role of alignment: the base model shows 1.0% single-cluster rate vs. 28.5% for the instruct model (p < 10^{-6}). A training stage ablation (Base 0.0% -> SFT 1.5% -> DPO 4.0% SCR) localizes the cause to DPO, not SFT. Cross-family replication on four model families reveals alignment tax severity varies by family and scale. We validate across 22 experiments, 5 benchmarks, 4 model families, and 3 model scales (3B-14B), with Jaccard, embedding, and NLI-based baselines at three DeBERTa scales (all ~0.51 AUROC). Cross-embedder validation with two independent embedding families rules out coupling bias. Cross-dataset validation on WebQuestions (58.0% SCR) confirms generalization beyond TruthfulQA. The central finding -- response homogenization -- is implementation-independent and label-free. Motivated by this diagnosis, we explore a cheapest-first cascade (UCBD) over orthogonal uncertainty signals. Selective prediction raises GSM8K accuracy from 84.4% to 93.2% at 50% coverage; weakly dependent boundaries (|r| <= 0.12) enable 57% cost savings.
The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州));Alibaba Group(阿里巴巴集团);Guangdong Laboratory of Artificial Intelligence and Digital Economy (Shenzhen)(广东人工智能与数字经济实验室(深圳))
Hierarchical Prompting Taxonomy: A Universal Evaluation Framework for Large Language Models Aligned with Human Cognitive Principles
Indian Institute of Technology Kharagpur(印度理工学院Khargapur分校);Rochester Institute of Technology(罗切斯特理工学院);Researcher, Fatima Fellowship(研究员,Fatima fellowship);AI Institute, University of South Carolina(人工智能研究所,南卡罗来纳大学);Amazon GenAI(亚马逊生成人工智能);Stanford University(斯坦福大学);James Silberrad Brown Center for AI, San Diego State University(詹姆斯·西伯拉德·布朗人工智能中心,圣地亚哥州立大学)
Instructional Goal-Aligned Question Generation for Student Evaluation in Virtual Lab Settings: How Closely Do LLMs Actually Align?
Department of EdPsych, University of Wisconsin-Madison(威斯康星大学麦迪逊分校教育心理学系);Department of CS, Wichita State University(威斯康星州立大学Wichita分校计算机科学系);Department of CSSE, Auburn University(阿伯茨罕大学计算机科学与工程系)
When AI Fails, What Works? A Data-Driven Taxonomy of Real-World AI Risk Mitigation Strategies
Faculty of Computer Science and Engineering(计算机科学与工程系);Ss. Cyril and Methodius University(西里尔和方法ius大学);Department of Computer Science(计算机科学系);Metropolitan College(Metropolitan学院)
IaC Generation with LLMs: An Error Taxonomy and A Study on Configuration Knowledge Injection
Jheronimus Academy of Data Science(Jheronimus数据科学学院);Tilburg University(蒂尔堡大学);Eindhoven University of Technology(埃因霍温理工大学);University of Sannio(萨诺尼大学)
WorldView-Bench: A Benchmark for Evaluating Global Cultural Perspectives in Large Language Models
Information Technology University Department of Computer Science(信息科技大学计算机科学系);Zayed University College of Interdisciplinary Studies(扎耶德大学跨学科研究学院);Computer Science and Engineering Department(计算机科学与工程系)
FinAuditing: A Financial Taxonomy-Structured Multi-Document Benchmark for Evaluating LLMs
Columbia University(哥伦比亚大学);Stevens Institute of Technology(史蒂文斯理工学院);Rensselaer Polytechnic Institute(拉特格斯理工学院);University of Montreal(蒙特利尔大学);McGill University(麦吉尔大学);MBZUAI(麦吉尔大学人工智能研究所);Cardiff University(卡迪夫大学);The University of Manchester(曼彻斯特大学);Harvard University(哈佛大学)
PathMem: Toward Cognition-Aligned Memory Transformation for Pathology MLLMs
University of Science and Technology of China(中国科学技术大学);Shenzhen University(深圳大学);Nanyang Technological University(南洋理工大学);Imperial College London(伦敦帝国学院);Huazhong University of Science and Technology(华中科技大学)
Quantification of Large Language Model Distillation
Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences(中国科学院深圳先进技术研究院);Peking University(北京大学);SUSTech(南方科技大学);SUAT;Leibowitz AI;UNSW Sydney(新南威尔士大学悉尼分校)
FaithSteer-BENCH: A Deployment-Aligned Stress-Testing Benchmark for Inference-Time Steering
University of Electronic Science and Technology of China(电子科技大学);Mohamed bin Zayed University of Artificial Intelligence(马尔代夫人工智能大学);South China University of Technology(华南理工大学)
Gaoling School of Artificial Intelligence, Renmin University of China(中国人民大学高瓴人工智能学院);Department of Computer Science and Technology, Tsinghua University(清华大学计算机科学与技术系);Tencent Inc.(腾讯公司)
EiCAP: Beyond Fluency, Probing and Improving Emotional Intelligence in LLMs via Psychologically Grounded Multi-Turn Dialogue
Qatar Computing Research Institute (QCRI), Hamad Bin Khalifa University(卡塔尔计算研究所(QCRI),哈马德·本·卡伊夫大学);University of Illinois Urbana-Champaign (UIUC)(伊利诺伊大学厄巴纳-香槟分校)