arXiv:2607. 05679v1 Announce Type: cross Abstract: Language models (LMs) exhibit problematic biases, such as stereotypes.
arXiv:2608. 05726v1 Announce Type: cross Abstract: Large Language Models (LLMs) are often used as evaluators of text quality, known as LLM-as-a-Judge, which can outperform conventional automatic evaluation metrics that rely on reference texts.
The paper introduces a unified framework that simultaneously measures intrinsic (encoded) and extrinsic (expressed) gender bias in large language models using identical neutral prompts. It finds a consistent link between latent gender information and output bias, but shows that alignment via supervised fine‑tuning reduces expressed bias while leaving internal gender associations largely intact and reactivatable by adversarial prompts. The study also demonstrates that debiasing gains on structured benchmarks may not transfer to realistic tasks such as story generation.
arXiv:2606. 08417v1 Announce Type: cross Abstract: Diffusion and continuous flow-based language models have emerged as the leading non-autoregressive alternatives to language modeling.
Large Language Models (LLMs) are often used as evaluators of text quality, known as LLM-as-a-Judge, which can outperform conventional automatic evaluation metrics that rely on reference texts. However, LLM evaluators tend to generate particular scores regardless of the context of the evaluated text, which is known as scoring bias.
arXiv:2606. 07537v1 Announce Type: cross Abstract: Large language models hallucinate--producing fluent, confident, factually wrong outputs--with a consistency that persists across generations and scales.
arXiv:2609.34240v2 Announce Type: replace-cross Abstract: Existing open-ended generation metrics measure likelihood, lexical diversity, or distributional similarity in generic representation space, y...
The paper examines whether removing declared language fields from de‑identified résumés eliminates demographic leakage in large language models. By keeping language attributes identical and varying only unstructured prose across five ethnocultural groups and three cue‑salience levels, the authors find that non‑language text still allows target‑group recovery (average 0.757, reaching 1.000 under high salience). They also show that evaluation design—such as allowing or forbidding ties—dramatically affects LLM‑as‑a‑judge outcomes, underscoring the importance of evaluation protocol in bias audits.
BiasGym is a cost‑effective, generalizable framework that injects specific biases into large language models via token‑based fine‑tuning while keeping the model frozen. It then uses two debiasing methods—Scope and Steer—to identify and suppress or redirect the components responsible for biased behavior. The framework enables consistent bias elicitation, precise localization of bias associations, and targeted debiasing without harming downstream performance, and it has been shown to reduce real‑world stereotypes such as labeling Italians as reckless drivers.
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.
The paper investigates whether adding Abstract Meaning Representation (AMR) data to large language models (LLMs) improves performance on downstream tasks. By reproducing recent studies and applying a consistent hyperparameter protocol, the authors find that text-only baselines match or surpass AMR-augmented models. A perplexity-based probe shows that AMR does not provide LLMs with additional relational knowledge, suggesting no clear benefit from AMR augmentation.
arXiv:2603.18908v5 Announce Type: replace Abstract: Independently trained language models often learn compatible late-stage representations, despite differences in training objectives, architectures,...
The paper introduces a bias depth score to differentiate between stable model preferences (Deep biases) and prompt‑dependent responses (Shallow biases) in large language models. By analyzing 4,442 opinion prompts across four models, it finds that only about a quarter of concentrated preferences persist after scenario reframing, indicating that most are shallow. The study shows Deep biases are more often inherited from pretraining and harder to remove through fine‑tuning or prompt‑based debiasing, highlighting the need to distinguish learned biases from prompt artifacts.