It is proposed that persona-based evaluation can serve as a scalable diagnostic of what generative systems value and prioritize when depicting humanity, and that persona generations are far from neutral.
Abstract
Synthetic data generated by large language models plays a central role in the training and alignment process of other AI systems. However, this process also risks inheriting the structural biases of organic corpora and embedding new biases that stem from the design choices underlying the data creation process. This paper examines the systematic biases that emerge when large language models (LLMs) are tasked with generating synthetic personas. We introduce a reproducible, minimally conditioned pipeline that produced 40,000 personas, in four different languages, using two instruction-tuned open-weight generators (Llama−3.3-70B-Instruct and Qwen2.5-72B-Instruct), followed by a battery of quantitative analyses: name match-rates, KL divergence/skew for gender, age-pyramid comparisons, profession-gender intersectionals, adjective/sentiment profiling, and Proppian role classification. Our main findings reveal that persona generations are far from neutral. Models tend to focus on middle-aged, aspirational, and overwhelmingly positive (i.e., upbeat/optimistic) characters, while non-binary identities and many real-world occupations remain underrepresented. We conclude that contemporary training and alignment regimes produce a form of narrative sanitization that both flattens representational diversity and embeds normative assumptions, and propose that persona-based evaluation can serve as a scalable diagnostic of what generative systems value and prioritize when depicting humanity.
Benevolence bias is identified and measure, a small but consistent tendency for aligned LLMs to lean toward the kinder, safer, more socially approved answer on value-laden survey questions, and is easy to diagnose and straightforward to fix.
Yuanzi Li, Jun-Hao Wang, Minghui Liu et al.· 0 citations
Language models (LMs) often acquire various biases during pre-training and may express them in interactions, potentially causing social harm. Existing methods often rely on counterfactual augmentation or representation projection. These strategies remain limited in practice due to their high computational costs and difficulty in scaling to larger models. Additionally, many of these strategies require manual data annotation, narrowing their scope to specific cultures and bias categories. To overcome these limitations, we propose HEIMAT, a HEurIstic-style autoMATic debiasing framework for LMs. HEIMAT consists of two main steps: bias disclosure and debiasing fine-tuning. In the first step, it uses simple templates to construct heuristic prompts, which are applied to reveal model biases and generate corresponding context prompts. In the second step, it fine-tunes the model by minimizing the Jensen-Shannon divergence of predictions on these context prompts to reduce bias. Extensive experiments show that HEIMAT effectively mitigates bias in different cultures while maintaining the model's natural language understanding (NLU) performance.
Tian Lan, Yemin Wang, Chuancheng Shi et al.· 0 citations
Evaluating 13-21 models across six presentation operationalizations and four task-domain operationalizations suggests that, despite confounds, some models possess practical SGTR capabilities, and that SGTR should be monitored and considered in the design of safety-critical AI applications.
J. St-Amand, Callum Canavan, S. Imran et al.· 0 citations
This work finds that likelihood-trained TPMs can result in failed generations due to overly large corrections to the LM’s logits, and trains TPMs with LM-aligned objectives that better align with the LM token-probability space.
Hanzhang Liu, William Zhao, Zilei Shao et al.· 0 citations
Detecting equivocation is essential, as indirect or evasive responses can shape public perception, influence political narratives, and undermine transparency in democratic discourse. To address the challenge of detecting evasive political responses on digital platforms, participation in the CLARITY SemEval-2026 Task was undertaken, which focuses on (i) clarity-level classification and (ii) fine-grained evasion-type classification in political question-answer contexts. This study introduces a data-centric framework that systematically examines the effects of class distribution and refinement strategies on the performance of Large Language Models (LLMs). A distribution-aware, LLM-augmented dataset was constructed by selectively paraphrasing minority-class instances to enhance class balance, and its performance was benchmarked against full, rebalanced, and undersampled training configurations. To comprehensively assess the proposed method, Qwen3-14B, Phi-4, Gemma-2 9B, and Mistral 7B were evaluated in in-context learning (ICL) settings (zero-shot and few-shot) and with LoRA fine-tuning. Experimental results indicate that fine-tuning Phi-4 with class rebalancing yields strong performance, achieving 74.77% on Subtask-1 and 51.55% on Subtask-2
A lot of research attention has been devoted to checking whether large language models (LLMs) are politically biased. This work has largely focused on high-level ideological dimensions, such as left--right or progressive--conservative, and it has been shown that while LLMs are predominantly left and progressive leaning, largely mimicking the biases in the training data, they can be to some extent steered to change their preferences in post-training. In this short note, we check if LLMs have robust stances with regard to major substantive societal issues, on which members of the same ideological camp are often in disagreement, summarised in a novel dataset \textsc{HardChoices}. We show that, faced with this line of questioning, LLMs, both large and small, surprisingly rarely declare neutrality, are often incoherent, and demonstrate a remarkable degree of agreement on issues where they do take stances.
