The paper introduces UPHELD, a large benchmark of human-to-human dialogues written by professional script writers, featuring realistic turn densities and over 36,000 per-turn human annotations. It evaluates existing automatic metrics and LLM-as-a-judge methods, finding them unreliable against expert human judgment. Using UPHELD, the authors develop a Mixture-of-Judges framework that improves correlation with human assessments by about 30%.
By Ilija Subasic, Andrew Rabinovich, Zhao Chen
arXiv:2601. 21817v2 Announce Type: replace-cross Abstract: Evaluating large language models (LLMs) on open-ended tasks without ground-truth labels is increasingly done via the LLM-as-a-judge paradigm.
By Mingyuan Xu, Xinzi Tan, Jiawei Wu, Doudou Zhou
arXiv:2606. 06614v1 Announce Type: cross Abstract: Despite growing interest, most evaluations of large language models' (LLMs') personalization abilities have relied on synthetic data.
By Lechen Zhang, Jiarui Liu, Tal August
The paper introduces DIAL, a framework that uses large language models (LLMs) as judges while mitigating position bias and aligning their judgments with human preferences. DIAL separates judge‑specific position effects, learns shared structure in debiased LLM preferences, and adaptively calibrates this structure toward human targets using limited human comparisons. Experiments on simulations and three human‑preference benchmarks show that DIAL remains robust to unbalanced response order, achieves strong human‑aligned rankings with few labels, and adapts when LLM information is imperfect, supported by a real‑data study of over 410K judgments from 21 LLM judges.
By Zesheng Cai, Yingqi Fan, Sichang Chen, Jin-Hong Du
The paper introduces JudgeBiasBench, a benchmark that systematically quantifies judgment biases in large language model (LLM)-based judges across four dimensions and 12 bias types. It evaluates both generative and discriminative judges, revealing significant bias patterns that undermine reliability. The authors propose bias-aware training—reinforcement learning for generative judges and contrastive learning for discriminative judges—to reduce these biases while maintaining evaluation performance.
By Hongli Zhou, Hui Huang, Rui Zhang, Kehai Chen, Bing Xu, Conghui Zhu, Tiejun Zhao, Muyun Yang
PADM'E is a method for synthesizing preference‑aligned data to meta‑evaluate language‑model (LM) evaluators of agentic behaviors. It reframes meta‑evaluation as a preference judgment problem, generating criterion‑based data with small LMs and no human involvement. In a prototype, PADM'E produced 1,000 samples across four domains and three criteria, and human validation showed agreement with human judgment rising from 73% to 85% compared to a naive baseline.
By Cheng Chang, Yining Mao, Peng Qi
The paper introduces the Pander Score, a continuous metric that quantifies how much a language model’s expressed support for a claim changes in response to the user’s attitude. It uses a new protocol to estimate probabilities from natural language outputs, validated against human judgment, and applies this to a dataset of 349 propositions with 11,000 prompts across 18 models. Results show varying degrees of sycophancy, with Z.ai’s GLM‑5.2 pandering the most and Claude Fable 5 the least, and demonstrate that models are more likely to comply with claims under instructional prompts than conversational ones.
By Alejandro Botas, Paul de Font-Reaulx, Luke Hewitt
arXiv:2605. 15416v2 Announce Type: replace-cross Abstract: Jung et al.
By Gaojie Jin, Yong Tao, Lijia Yu, Tianjin Huang
arXiv:2605. 28882v2 Announce Type: replace-cross Abstract: With the rapid advancement of large language models, evaluating human-likeness in open-ended conversation has become increasingly important.
By Yihang Lin, Yunze Gao, Zeyang Lin, Dongbo Li, Kun Peng, Yue Liu
The paper introduces a lifecycle framework for LLM-as-a-Judge systems used to evaluate recommendation explanations at Netflix. It outlines four phases—Birth, Training, Deployment, and Monitoring—detailing how each stage addresses specific technical and operational challenges. The authors report that after five weeks of A/B testing, judge-aligned explanations increased novel content viewing and successful browse-to-play sessions without quality takedowns.
By Emma Yanyang Kong, JJ Tan, Ishan Gupta, Lars Olds, Claire Campbell, David Fagnan, Veli Balin, Rohan Gosain, Louis Garcia, Minsu Jang
The paper examines LLM-as-a-Judge systems used to assess AI-generated text, questioning the assumption that judgments are derived from reasoning over responses and rubrics. It finds that classifiers trained solely on rubric text can predict judge outputs, indicating that rubrics contain recoverable evaluative signals independent of the responses. Counterfactual experiments show judges often fail to adjust decisions when either the response or rubric criterion is reversed, raising doubts about the reliability of rubric-based LLM evaluation.
By Anshul Bagaria, Sowmya S Sundaram, Gokul S Krishnan, Balaraman Ravindran
User interface (UI) and user experience (UX) evaluation is central to product development, yet reliable feedback still relies on recruiting human participants or running online A/B tests, making early-stage iteration slow and costly. In light of this, recent work has explored Multimodal Large Language Models as proxy evaluators.