The paper investigates whether consensus among large language model (LLM) judges truly reflects human alignment. By treating each judge’s scores as vectors, the authors measure spread, effective rank, and angles to human scores across 42 judges on Indic benchmarks, revealing that inter‑judge agreement often mirrors shared blind spots rather than human judgments. They find that while judges agree as much as humans, they only reach 58‑66% of human agreement and frequently focus on axes humans do not weight, indicating that ensemble agreement alone is insufficient evidence of alignment.
arXiv:2609.13824v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly used to evaluate the responses of other language models. This approach, known as LLM-as-a-Judge, is faste...
arXiv:2601.08654v3 Announce Type: replace Abstract: Rubric-based text evaluation increasingly relies on large language models (LLMs) as scalable judges, yet frozen black-box models can interpret the...
JudgeSense is a benchmark comprising 880 items from human‑labelled corpora, each presented under two differently worded instructions that ask the same question. The study evaluates 25 judges from six providers across four tasks, measuring how rewording affects agreement with the judge’s own verdicts. Results show that rewording reduces agreement on all tasks, with significant effects on two, and that stability varies across tasks and is not predicted by parameter count.
arXiv:2608.29517v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly used as essay graders in learning analytics, evaluated almost exclusively with agreement statistics. Educ...
The study evaluates a label‑free rubric‑clarification procedure that rewrites only the most contested sub‑criteria of a rubric, aiming to reduce disagreement among large‑language‑model (LLM) judges. Across six rubric conditions and three datasets (SummEval, HealthBench, USR Topical‑Chat), the method increased inter‑judge agreement on SummEval by 0.069 Krippendorff’s α, but did not improve agreement with human expert ratings. In other datasets, neither inter‑judge nor human agreement met the preregistered effect criteria, and human agreement was actually lower after editing contested clauses compared to random edits.
arXiv:2608.30373v1 Announce Type: new Abstract: Multi-Agent Debate (MAD) has been widely adopted to improve LLM-based evaluation by prompting multiple agents to negotiate and reach a consensus. Howev...
Large Language Model judges are commonly used to rank texts via pairwise comparison, with reliability traditionally measured by position bias, transitivity, and pairwise agreement. This paper argues that these proxies are misleading because they are dominated by close‑rank‑gap pairs, which contribute little to the overall ranking, while far‑gap pairs carry the true ranking signal. Experiments on simulations and human‑rated corpora show weak correlation between the proxies and actual ranking accuracy, suggesting judges should be evaluated using rank‑gap‑conditional metrics against human rankings.
Multi-Agent Debate (MAD) has been widely adopted to improve LLM-based evaluation by prompting multiple agents to negotiate and reach a consensus. However, for subjective rubric-based scoring, inter-ag...
Large language models (LLMs) are widely used as automatic judges, with validity typically assessed via alignment with human scores. However, aggregate agreement fails to reveal whether humans and LLMs...
The paper investigates how well large language models (LLMs) can serve as judges in summarization evaluation by applying psychometric techniques. Using Many‑Facet Rasch Models, the authors decompose human and LLM ratings into latent summary quality, rater severity, dimension severity, and rating‑scale thresholds, and introduce a residual hardness metric to capture judging difficulty. Their analysis of 17 open‑weight LLM judges on the SummEval dataset reveals that moderate alignment in latent quality does not translate to alignment in residual hardness; humans and LLMs differ in which summary–dimension units are hard, with LLMs tending to find consistency hard and humans tending to find coherence hard, and some hard cases can be predicted from observable source–summary properties.
The paper demonstrates that reading a large language model (LLM) judge’s verdict from the logits of its first generated token—an approach used in constrained decoding and likelihood‑scoring evaluation—introduces a significant distortion in position bias. Because judges do not always start with a verdict token (12–49% of cases for Qwen3 judges and <3% for Llama‑3.1‑8B and Phi‑3.5‑mini), this readout often returns the first response rather than a true judgment, inflating position bias by up to 42 points while barely affecting judge accuracy. The authors recommend reporting the frequency with which a judge leads with a verdict token to provide a more accurate assessment of position bias.
arXiv:2609.21277v1 Announce Type: cross Abstract: How many human judgments does a panel of language models represent? The answer depends on what is matched. We audit categorical judge panels against...