Prior work on human label variation (HLV) in natural language inference (NLI) has often relied on re-annotation resources that select items by disagreement level. An earlier study (arXiv:2607.
arXiv:2607.15870v2 Announce Type: replace Abstract: Human label variation in natural language inference is increasingly treated as signal rather than noise, but how much of it formal semantic structu...
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...
arXiv:2606. 28772v1 Announce Type: cross Abstract: Hate speech annotation pipelines routinely collapse annotator disagreement into majority vote labels before training.
arXiv:2608. 11138v1 Announce Type: cross Abstract: We propose that a model's uncertainty about a token is reflected not only in the breadth of its output distribution but also in whether a confident prediction is \emph{fragile} under perturbation of its attention pathways.
Warning: This paper studies stereotypes and biases, and contains potentially disturbing examples, used for illustration purposes only. Our findings should not be interpreted as an argument against alignment.
The study investigates bias in large language model (LLM) judges by having ten LLMs evaluate narrative constraint selections rather than generated text. Results show that self-preference largely disappears under blind evaluation when quality and evaluator severity are controlled, but self- and other-labels alone shift scores bidirectionally when quality is matched. The authors conclude that authorship attribution drives evaluation bias and that open-ended, ground‑truth‑free tasks can effectively study LLM judge behavior.
arXiv:2609.38448v1 Announce Type: new Abstract: Open-ended LLM homogeneity can create false plurality when several systems appear to offer independent perspectives while returning the same familiar d...
arXiv:2501. 02211v3 Announce Type: replace-cross Abstract: Large language models (LLMs) reproduce homogeneity bias -- the tendency to portray marginalized groups as more internally similar than dominant groups -- but whether this bias generalizes across models, is stable under different inference settings, or depends on how group identity is signaled remains unstudied.
The study investigates how different prompt components affect language model responses in psychometric tests. By crossing five distinct baseline personas with five variants of each prompt element—persona wording, task instruction, item wording, and option symbol—the authors measure response shifts using the 1‑Wasserstein distance. Their analysis of 13 small open‑weight language models on the Big Five Inventory and Short Dark Triad reveals that task instruction and option symbol changes often cause more variation than paraphrasing the persona or item, with prompt artifacts explaining over 50% of the variation for many items.
arXiv:2609.37616v1 Announce Type: new Abstract: Language models tend to agree with whatever a user asserts, and post-training increasingly targets this sycophancy so that models evaluate claims on th...
arXiv:2606. 05970v1 Announce Type: cross Abstract: Large language models are increasingly used for structured extraction from clinical free-text notes, but the sensitivity of their output to upstream configuration choices is less understood than their accuracy on fixed benchmarks.
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.