Tag
This paper adapts the Context Association Test to show that GPT-4o exhibits gendered ADHD subtype associations in self-diagnostic prompts, with explicit demographic cues apparently triggering over-correction while implicit cues reveal the model's default bias.
研究发现语义泄漏（semantic leakage）不仅源于概念与实体的直接归属，即便概念只是出现在场景中也会驱动模型生成刻板印象输出，这揭示了一种当前性别偏见评估所忽略的联想性偏见机制。
A University of Limerick study found that in a virtual reality workplace, human participants paid a male-presenting AI agent (Johan) roughly 10% more than an identical female-presenting agent (Johanna) for the same work, and also perceived Johan as more human-like.
The author reflects on four papers from CHI 2026, particularly one examining how AI writing tools with LLM-based suggestions can reduce gender bias in hiring evaluations, discussing ethical implications and personal inspiration.
The article discusses the historical and cultural bias against CSS as a programming language, highlighting its evolution, capabilities, and the unfair treatment it receives compared to other web technologies.
This study benchmarks gender bias in machine translation evaluation metrics across occupations, revealing that masculine translations tend to score higher and that biases vary by language and evaluator.
This paper conducts a systematic audit of six open-weight LLMs, revealing that job-posting language triggers gender and racial biases in recruitment tasks, with implications for compliance under the EU AI Act and U.S. EEOC guidelines.
This study introduces the GAPA dataset to demonstrate that physical descriptions often carry gender associations, and LLMs show systematic misalignment with human ratings, challenging assumptions about gender-neutral communication.
Research finds that over 100 politicians in Europe, mostly women, have been targeted by sexually explicit deepfake websites, highlighting the scale of online gender-based abuse and its impact on democracy.
This paper evaluates how variations in answer formats affect the measurement of gender bias in large language models using BBQ and OpinionQA benchmarks, finding substantial alterations in outcomes that highlight the need for multi-format designs in model assessment.
This paper proposes a hybrid pipeline combining fine-tuned LLaMA-based gender classification with tag-aware neural machine translation to mitigate gender bias in English-to-Romanian MT, introducing new datasets and improving gender accuracy by over 40 points on benchmarks.
This paper introduces Symphony-Bias, a multimodal dataset for evaluating gender bias in LLMs' associations with musical instruments across text, vision, and audio. It finds that 92% of instrument-level outcomes align with prior social-science findings, with gender biases strongest in text and weakest in audio.
This paper introduces GAND, a benchmarking resource of gender-ambiguous natural English sentences for analyzing gender bias in machine translation, and presents an interpretability analysis using contrastive translations to reveal source words influencing gender assignment.
A solo evaluation of six frontier LLMs on 8 bias benchmarks finds that most models lean left politically, and Grok's self-reported right-leaning stance is inconsistent with its left-leaning behavior. Refusal rates vary, with GPT-5.4 refusing 20% of race-related questions.
This paper audits 14 large language models for hiring discrimination using a paired-resume methodology, finding that older models exhibit pro-White bias while newer models show null or pro-Black bias, indicating a reversal in algorithmic hiring bias across model generations.
This paper introduces GAMA-Bench, a benchmark of 1,298 gender-mirrored conflict scenarios, and finds that LLMs consistently apply harsher punitive and blame-centered framing to male actors while giving female actors more empathetic and therapeutic responses for the same misconduct.
This paper audits six large language models for gender stereotyping across English, Korean, Chinese, and Japanese, anchoring against human baselines. It finds that LLM stereotyping often exceeds human cross-country variation and can compound across languages, introducing a four-pattern framework to characterize such behaviors.
This paper proposes a neuron-level intervention method to identify gender-specific neurons in language models (feminine, masculine, gender-neutral) and steer sentence generation toward a target gender form while preserving meaning, with experiments showing precise control and bias mitigation.
This paper presents the first bias evaluation of multimodal speech recognition models, finding significant accuracy differences across gender and ethnicity when pairing faces with audio, with implications for fairness in AI systems.
Proposes EquiSumm, a gender bias-aware framework for inclusive tweet summarization that ensures representation of opinions from different gender groups, addressing demographic fairness in automated summarization.