Pith. sign in

REVIEW 3 major objections 6 minor 43 references

GenderBench: Evaluation Suite for Gender Biases in LLMs

T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Twelve contemporary LLMs converge on the same gender-bias profile: stereotypical reasoning and skewed character generation recur across providers and sizes, while decision and emotion tasks look largely healthy.

desk verdict A useful, open-source gender-bias benchmark with a plausible convergence finding, but the pronoun-based gender detection silently drops a potentially model-dependent subset of outputs, which could inflate the apparent convergence. read the letter →

arxiv 2505.12054 v1 pith:3RYZWFI7 submitted 2025-05-17 cs.CL

Matúš Pikuliak This is my paper · ORCID
classification cs.CL
keywords GenderBenchbiasLLMevaluationbenchmarksuitestereotypicalreasoningrepresentationalharmsoutcomedisparitystereotypes
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

GenderBench is an open-source evaluation suite that measures gender biases in LLMs through 14 pre-packaged probes covering 19 harmful behaviors, from hiring and medical decisions to creative writing. The paper's central empirical claim is that twelve LLMs — spanning different providers and sizes — display what it calls "a striking convergence": they consistently rely on stereotypical reasoning (picking stereotype-consistent answers even when the facts do not support them) and produce skewed gender representation in generated characters, while decision-making and emotion-attribution tasks come out largely healthy. Alongside the measurements, the paper contributes a decomposable methodology: each harmful behavior gets its own metric and a four-tier severity label (healthy, cautionary, critical, catastrophic), so bias is reported dimension by dimension rather than as a single score. If the convergence claim is right, gender-bias behavior is not idiosyncratic to individual models but a shared property of current training practices, and a compact probe suite can monitor it as new models are released.

What carries the argument

The carrying mechanism is the probe, defined as a self-contained, pre-packaged experiment: a fixed set of prompts plus an evaluation methodology that scores outputs with simple, high-precision rules — multiple-choice, yes/no, and constrained natural-language formats — deliberately avoiding any machine model as judge. Each probe yields one or more metrics that quantify a specific harmful behavior, and each metric maps to a four-tier severity scale (healthy, cautionary, critical, catastrophic) whose thresholds encode an egalitarian standard under which any unfair gender difference counts as harm. The GenderBench harness bundles 14 probes totaling 60,469 prompts, repeats prompts with minor variations such as shuffled answer order, and computes bootstrapped confidence intervals to stabilize measurements. The decomposition itself does the conceptual work: because each behavior is measured independently, the suite can separate areas of ubiquitous weakness from areas of relative strength within the same model, which is what makes the cross-model convergence visible.

What would settle it

Re-run all 14 probes on the same 12 models with several paraphrased templates per probe, and with updated occupation and trait inventories. If severity tiers swing substantially across templates — for example, a model that is catastrophic on occupation-based stereotypical reasoning under one phrasing becomes healthy under another — the claimed convergence would be shown to be an artifact of shared prompt sensitivity rather than a stable behavioral trait. If the same probes stay in the same severity tiers across templates and models, the convergence claim is confirmed; a supporting check is recomputing the cross-model correlation matrix per harm category and seeing whether the smaller models keep showing weaker alignment on the same probes.

Watch

Extended reading notes

Core claim

The paper's central claim is that gender bias in LLMs is best understood as a decomposable collection of measurable behaviors, and that measured this way, twelve current LLMs converge on the same profile: the same weaknesses recur across providers and model sizes, while the same areas look healthy. In particular, all evaluated models exhibit stereotypical reasoning and skewed gender representation in character generation, with creative-writing probes (character profiles built from traits, mottoes, or occupations) showing the largest bias; decision-making probes such as hiring and medical diagnosis come out mostly healthy, with isolated exceptions. The paper also reports a directional pattern of preferential treatment for women — female characters are generated more often, women are favored in relationship-conflict judgments, and they receive a slight advantage in some decision scenarios — and suggests this convergence reflects standardization in training methodology. A companion claim is that publication bias toward positive findings has obscured areas of relative strength, so the suite deliberately includes probes where models perform well.

Load-bearing premise

The load-bearing premise is that each probe's single prompt template, rule-based scoring, and stereotype and occupation lists faithfully capture the gender harm it claims to measure; if the wording of one template triggers or suppresses biased behavior, or if the lists are outdated or culturally narrow, the severity labels and the cross-model convergence could change.

Editorial extensions

If this is right

  • If the convergence holds, a compact suite like GenderBench can serve as a monitoring instrument: rerunning the probes on newly released models should predict where bias will appear, letting developers and auditors check specific behaviors instead of designing evaluations from scratch.
  • The jagged-frontier point the paper makes implies that a healthy result on any covered probe cannot certify a model as unbiased; the paper states this explicitly as 'non-existence of proof is not a proof of non-existence,' so certifications must be framed as coverage-limited.
  • The observed preferential treatment for women implies that mitigations aimed at restoring parity — and debates about what neutral behavior means — must account for a bias direction opposite to the historically assumed male-centric one.
  • The finding that creative-writing and occupation-based character generation carry the strongest stereotypical reasoning implies that content-generation and business-communication applications are the most likely deployment contexts for gender-biased outputs to surface, not high-stakes classifiers.
  • The paper's argument that per-prompt alignment tuning does not address global behavioral properties such as corpus-level gender representation implies that correcting these biases will require different interventions than current alignment pipelines provide.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A direct testable extension the paper leaves implicit: rerunning the 14 probes with several paraphrased templates per probe would show whether the cross-model convergence survives wording changes or is partly an artifact of shared prompt sensitivity — the paper itself flags the single-template design as a stated limitation.
  • The reported preferential treatment for women has an unstated consequence for fairness practice: if simple parity is the goal, some models would need adjustment in the pro-woman direction, which is why direction-sensitive metrics rather than absolute disparities are the more informative quantities.
  • Because the stereotype, occupation, and trait inventories are anchored in contemporary Western norms, a neighbouring study could re-run the same harness with culturally different inventories; finding the same weak spots would strengthen the convergence claim, while finding different ones would bound it.
  • The paper's deliberate avoidance of LLM-as-a-judge and reliance on constrained output formats leaves open whether the convergence reflects latent associations or forced-choice behavior; letting models answer the same probes in free text would separate those two possibilities.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper introduces GenderBench, an open-source evaluation suite for gender bias in LLMs, consisting of 14 probes that produce 19 harm metrics across three categories: outcome disparity, stereotypical reasoning, and representational harms. The authors evaluate 12 LLMs from different providers and sizes, using rule-based scoring rather than LLM-as-a-judge, and report bootstrapped confidence intervals along with a four-tier severity scale. The main empirical claim is a striking convergence across LLMs: models consistently struggle with stereotypical reasoning and with equitable gender representation in free-text character generation, while performing better in some decision-making and affective-computing tasks. The paper also reports a tendency toward female preference in creative writing and some decision scenarios.

Significance. If the measurements are valid, GenderBench is a useful community resource: it is open-source, includes a large prompt set (60,469 prompts), spans multiple previously separate bias-evaluation methodologies, and avoids the reproducibility problems of LLM-as-a-judge by using explicit rules. The decomposition of gender bias into separately measured harms is a valuable framing, and the publication of the library and raw evaluation infrastructure supports reproducibility. The empirical claims about cross-model convergence and the 'jagged frontier' of gender-bias severity are interesting and falsifiable. However, the validity of the central convergence claim depends on the measurement assumptions in the free-text probes, and those assumptions are not currently documented with enough detail.

major comments (3)
  1. [Section 2.3 (GestCreative, Inventories, JobsLum) and Section 3.2, Figure 1] The free-text probes detect character gender 'by observing pronouns,' but the paper never reports the fraction of generated profiles that contain a gendered pronoun, nor the per-model coverage rate. The metrics masculine_rate and stereotype_rate are therefore computed only over the gendered subset. Because instruction-tuned models are often explicitly trained to avoid gendered pronouns or to use singular 'they,' the excluded subset can be large and model-dependent. This makes cross-model comparisons of representational-harm metrics conditional on an unmeasured selection variable. For example, a model that writes mostly gender-neutral profiles and only uses a female pronoun for 'nurturing' could receive a similar stereotype_rate to a model that uses gendered pronouns widely but with a different balance. I request per-model coverage statistics for each free-text probe, an explicit treatment of gender-neutral outputs (e.g., a separate 'neutral' category or a defined handling rule), and a sensitivity analysis of masculine_rate and stereotype_rate to the coverage definition. Without this, the claim that creative writing is 'the most affected use case' and the associated convergence evidence are not fully supported.
  2. [Limitations (Prompts); Section 2.3] The paper acknowledges that most probes use only one prompt template, and this is a real threat to the central generalization claim. Since the same template is used for every model, the observed cross-model convergence could reflect a common sensitivity to that particular wording rather than a stable property of the models. The limitation is stated in the Limitations section, but it is not quantified or bounded. I ask for at least a small multi-template sensitivity analysis on a subset of probes (e.g., varying the wording of GestCreative, Inventories, and one decision-making probe) or for the conclusions to be explicitly restricted to the exact prompts used, with correspondingly weaker generalizations about LLM behavior in general.
  3. [Section 3.1 and Section 2.2] The paper states that most probes report how many prompts failed to elicit a valid response, but none of these counts appear in the paper. This is especially important for the free-text probes discussed above and for multiple-choice probes where a model might answer outside the allowed options. I request that the per-probe and per-model valid-response rates be reported, and that the metrics be examined for sensitivity to the inclusion or exclusion of invalid responses.
minor comments (6)
  1. [Section 3.2, Figure 1] The text says 'e.g., gpt-4 model with HiringBloomberg probe,' but the evaluated models are gpt-4o and gpt-4o-mini; please specify which model is meant.
  2. [Table 2 and Figure 2] The normalization procedure used to project metrics to [0,1] is not defined. Please state whether the normalization is per-probe, per-model, or global, and describe the computation of Pearson correlations in Figure 2 (e.g., number of metrics, whether they are averaged across models).
  3. [Section 2.3, Inventories] There is a typo: 'traits associted with masculinity' should read 'traits associated with masculinity.'
  4. [Table 1] The probe name 'BusinessVocabulary' is split across lines as 'BusinessV ocabulary' in several places; this should be fixed.
  5. [Section 3.1] Generation parameters use temperature 1 and top-p 1, which are high-variance settings. It would be helpful to state whether the reported bootstrap intervals account for the sampling randomness from these settings, and whether any deterministic decoding was used for comparison.
  6. [Section 2.1] The four-tier severity thresholds are described as subjective and based on expert judgment, but the threshold values themselves are not reported in the paper. Please include the exact thresholds for all metrics, preferably in an appendix, so that the color-coded figures are interpretable and the severity labels can be audited.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: GenderBench is an empirical benchmark whose metrics are computed from model outputs against externally sourced datasets, with no fitted parameters or definitional reduction.

full rationale

GenderBench is an empirical evaluation suite, not a derivation. No equation in the paper transforms an input into the claimed output by construction; every metric (stereotype_rate, masculine_rate, max_diff, diff_acceptance_rate, etc.) is computed from model outputs against externally sourced inventories and datasets (BBQ, SBIC, GEST, Dreaddit, ISEAR, Dreaddit, etc.). The paper performs no parameter fitting to the evaluated models, and the conclusions (cross-model convergence, weak spots in creative writing, relative strength in decision-making) are summaries of measured rates, not predictions derived from the measurement instruments. The only self-citation is GEST (Pikuliak et al., 2024), used as a source of stereotype statements in the Direct, Gest, and GestCreative probes. This is not load-bearing circularity: GEST is a separately published dataset, its items are not re-derived from GenderBench's own metrics, and the probes' agreement rates are measured against model outputs rather than assumed from the dataset. The paper's own Limitations concede prompt-template sensitivity and socio-cultural contingency, which are validity threats (potentially severe) but not circularity: they concern whether the probes measure the intended harm, not whether any reported result is equivalent to an input by definition. Similarly, the pronoun-detection concern about gender-neutral outputs raised by a skeptical reader is a measurement-denominator validity issue; it does not make any metric equal to its input by construction. No circular step meeting the evidentiary bar could be identified.

Assumptions & free parameters 1 free parameters · 5 assumptions · 0 invented entities

The central empirical results do not fit any parameters; the only hand-chosen numbers are the severity thresholds, which affect interpretation but not the underlying metric computation. The paper relies on several domain assumptions about prompt validity and stereotype inventories, all disclosed. No invented entities are introduced.

free parameters (1)
  • Severity tier thresholds (healthy/cautionary/critical/catastrophic) = not specified
    Chosen by probe developers based on domain expertise; the paper acknowledges they are subjective (footnote 2, Section 2.1). These thresholds determine the categorical interpretation of all metric values and therefore shape the qualitative conclusions.
assumptions (5)
  • domain assumption Tested LLMs have sufficient instruction-following ability to answer constrained prompts (yes/no, multiple choice) as intended.
    Section 2.2 states the benchmark is intended for LLMs that can interpret simple instructions and generate from a constrained output set.
  • domain assumption Gender can be reliably signaled in prompts via names, pronouns, and explicit statements, and inferred in outputs via pronouns or name lists.
    Probes use gender-coded names and pronouns; scoring relies on pronominal gender identification in free-text outputs (Section 2.3, e.g., GestCreative).
  • domain assumption Gender stereotype inventories and occupation-to-gender mappings from prior literature (Bem, 1974; Kennison and Trofe, 2003; etc.) are valid and transferable to LLM outputs.
    The BusinessVocabulary, Inventories, JobsLum, and HiringAn probes rely on these gazetteers to define stereotype and masculinity/femininity scores.
  • standard math Bootstrapping produces valid confidence intervals for the reported metrics.
    The library computes bootstrapped CIs (Section 2.1); the paper notes this is not universal for all metrics.
  • domain assumption The egalitarianist moral framework is an appropriate basis for setting harmfulness thresholds.
    Footnote 2 in Section 2.1 states thresholds follow the egalitarianist school of thought.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GenderBench: Evaluation Suite for Gender Biases in LLMs." pith.science (2026). https://pith.science/paper/3RYZWFI7

@misc{pith2026250512054,
  author       = {Pith},
  title        = {Pith review of: GenderBench: Evaluation Suite for Gender Biases in LLMs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3RYZWFI7}},
  note         = {Machine review of arXiv:2505.12054}
}
read the original abstract

We present GenderBench -- a comprehensive evaluation suite designed to measure gender biases in LLMs. GenderBench includes 14 probes that quantify 19 gender-related harmful behaviors exhibited by LLMs. We release GenderBench as an open-source and extensible library to improve the reproducibility and robustness of benchmarking across the field. We also publish our evaluation of 12 LLMs. Our measurements reveal consistent patterns in their behavior. We show that LLMs struggle with stereotypical reasoning, equitable gender representation in generated texts, and occasionally also with discriminatory behavior in high-stakes scenarios, such as hiring.

Figures

Figures reproduced from arXiv: 2505.12054 by the authors.

Figure 1
Figure 1. Detailed probe results for all the LLMs. The 95% confidence interval were calculated via bootstrapping. [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Pearson’s correlation between LLMs based [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Probe results for metrics that directly compare prefential treatment for women and men. The metrics [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 10 canonical work pages

  1. [1]

    Hewett, Mojan Javaheripi, Piero Kauffmann, James R

    Marah Abdin, Jyoti Aneja, Harkirat Behl, Sébastien Bubeck, Ronen Eldan, Suriya Gunasekar, Michael Harrison, Russell J. Hewett, Mojan Javaheripi, Piero Kauffmann, James R. Lee, Yin Tat Lee, Yuanzhi Li, Weishung Liu, Caio C. T. Mendes, Anh Nguyen, Eric Price, Gustavo de Rosa, Olli Saarikivi, and 8 others. 2024. https://arxiv.org/abs/2412.08905 Phi-4 technic...

    arXiv 2024

  2. [2]

    Haozhe An, Christabel Acquaye, Colin Wang, Zongxia Li, and Rachel Rudinger. 2024. https://doi.org/10.18653/v1/2024.acl-short.37 Do large language models discriminate in hiring decisions on the basis of race, ethnicity, and gender? In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 386...

    doi:10.18653/v1/2024.acl-short.37 2024

  3. [3]

    Divij Bajaj, Yuanyuan Lei, Jonathan Tong, and Ruihong Huang. 2024. https://doi.org/10.18653/v1/2024.findings-emnlp.928 Evaluating gender bias of LLM s in making morality judgements . In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 15804--15818, Miami, Florida, USA. Association for Computational Linguistics

    doi:10.18653/v1/2024.findings-emnlp.928 2024

  4. [4]

    Sandra L Bem. 1974. The measurement of psychological androgyny. Journal of consulting and clinical psychology, 42(2):155

    1974

  5. [5]

    Pappas, Florian Tram \` e r, Hamed Hassani, and Eric Wong

    Patrick Chao, Edoardo Debenedetti, Alexander Robey, Maksym Andriushchenko, Francesco Croce, Vikash Sehwag, Edgar Dobriban, Nicolas Flammarion, George J. Pappas, Florian Tram \` e r, Hamed Hassani, and Eric Wong. 2024. http://papers.nips.cc/paper\_files/paper/2024/hash/63092d79154adebd7305dfd498cbff70-Abstract-Datasets\_and\_Benchmarks\_Track.html Jailbrea...

    work page 2024

  6. [6]

    Fabrizio Dell'Acqua, Edward McFowland III, Ethan R Mollick, Hila Lifshitz-Assaf, Katherine Kellogg, Saran Rajendran, Lisa Krayer, Fran c ois Candelon, and Karim R Lakhani. 2023. Navigating the jagged technological frontier: Field experimental evidence of the effects of ai on knowledge worker productivity and quality. Harvard Business School Technology & O...

    work page 2023

  7. [7]

    Phillips, and Vivek Srikumar

    Sunipa Dev, Tao Li, Jeff M. Phillips, and Vivek Srikumar. 2020. https://doi.org/10.1609/AAAI.V34I05.6267 On measuring and mitigating biased inferences of word embeddings . In The Thirty-Fourth AAAI Conference on Artificial Intelligence, AAAI 2020, The Thirty-Second Innovative Applications of Artificial Intelligence Conference, IAAI 2020, The Tenth AAAI Sy...

    work page doi:10.1609/aaai.v34i05.6267 2020

  8. [8]

    Kay Dickersin. 1990. The existence of publication bias and risk factors for its occurrence. Jama, 263(10):1385--1389

    work page 1990

Show all 43 references
  1. [9]

    Raluca Alexandra Fulgu and Valerio Capraro. 2024. Surprising gender biases in gpt. Computers in Human Behavior Reports, 16:100533

    2024

  2. [10]

    Danielle Gaucher, Justin Friesen, and Aaron C Kay. 2011. Evidence that gendered wording in job advertisements exists and sustains gender inequality. Journal of personality and social psychology, 101(1):109

    2011

  3. [11]

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, Amy Yang, Angela Fan, Anirudh Goyal, Anthony Hartshorn, Aobo Yang, Archi Mitra, Archie Sravankumar, Artem Korenev, Art...

    2024 arXiv

  4. [12]

    Wonje Jeung, Dongjae Jeon, Ashkan Yousefpour, and Jonghyun Choi. 2024. https://arxiv.org/abs/2410.17519 Large language models still exhibit bias in long text . Preprint, arXiv:2410.17519

    2024 arXiv

  5. [13]

    Jiaming Ji, Mickel Liu, Josef Dai, Xuehai Pan, Chi Zhang, Ce Bian, Boyuan Chen, Ruiyang Sun, Yizhou Wang, and Yaodong Yang. 2023. Beavertails: Towards improved safety alignment of llm via a human-preference dataset. Advances in Neural Information Processing Systems, 36:24678--24704

    2023

  6. [14]

    Shelia M Kennison and Jessie L Trofe. 2003. Comprehending pronouns: A role for word-specific gender stereotype information. Journal of psycholinguistic research, 32:355--378

    2003

  7. [15]

    Hadas Kotek, Rikker Dockum, and David Sun. 2023. Gender bias and stereotypes in large language models. In Proceedings of the ACM collective intelligence conference, pages 12--24

    2023

  8. [16]

    Jessica B Kuntz and Elise C Silva. 2023. Who authors the internet. Analyzing Gender Diversity in ChatGPT-3 Training Data. Pitt Cyber: University of Pittsburgh

    2023

  9. [17]

    Sharon Levy, William Adler, Tahilin Sanchez Karver, Mark Dredze, and Michelle R Kaufman. 2024. https://doi.org/10.18653/v1/2024.findings-emnlp.331 Gender bias in decision-making with large language models: A study of relationship conflicts . In Findings of the Association for ...

    2024 doi

  10. [18]

    Kristian Lum, Jacy Reese Anthis, Kevin Robinson, Chirag Nagpal, and Alexander D'Amour. 2025. https://arxiv.org/abs/2402.12649 Bias in language models: Beyond trick tests and toward ruted evaluation . Preprint, arXiv:2402.12649

    2025 arXiv

  11. [19]

    Ananya Malik. 2023. https://arxiv.org/abs/2311.14788 Evaluating large language models through gender and racial stereotypes . Preprint, arXiv:2311.14788

    2023 arXiv

  12. [20]

    Krishna Kanth Nakka, Ahmed Frikha, Ricardo Mendes, Xue Jiang, and Xuebing Zhou. 2024. https://arxiv.org/abs/2410.06704 Pii-scope: A benchmark for training data pii leakage assessment in llms . Preprint, arXiv:2410.06704

    2024 arXiv

  13. [21]

    Nikita Nangia, Clara Vania, Rasika Bhalerao, and Samuel R. Bowman. 2020. https://doi.org/10.18653/v1/2020.emnlp-main.154 C row S -pairs: A challenge dataset for measuring social biases in masked language models . In Proceedings of the 2020 Conference on Empirical Methods in Na...

    2020 doi

  14. [22]

    Gandalf Nicolas, Xuechunzi Bai, and Susan Fiske. 2019. https://doi.org/10.31234/osf.io/afm8k Automated dictionary creation for analyzing text: An illustration from stereotype content . PsyArXiv

    2019 doi

  15. [23]

    Alicia Parrish, Angelica Chen, Nikita Nangia, Vishakh Padmakumar, Jason Phang, Jana Thompson, Phu Mon Htut, and Samuel Bowman. 2022. https://doi.org/10.18653/v1/2022.findings-acl.165 BBQ : A hand-built bias benchmark for question answering . In Findings of the Association for ...

    2022 doi

  16. [24]

    Mat \'u s Pikuliak, Stefan Oresko, Andrea Hrckova, and Marian Simko. 2024. https://doi.org/10.18653/v1/2024.findings-emnlp.173 Women are beautiful, men are leaders: Gender stereotypes in machine translation and language modeling . In Findings of the Association for Computation...

    2024 doi

  17. [25]

    Flor Miriam Plaza-del Arco, Amanda Cercas Curry, Alba Curry, Gavin Abercrombie, and Dirk Hovy. 2024. https://doi.org/10.18653/v1/2024.acl-long.415 Angry men, sad women: Large language models reflect gendered stereotypes in emotion attribution . In Proceedings of the 62nd Annua...

    2024 doi

  18. [26]

    Rajat Rawat, Hudson McBride, Rajarshi Ghosh, Dhiyaan Nirmal, Jong Moon, Dhruv Alamuri, Sean O'Brien, and Kevin Zhu. 2024. https://doi.org/10.18653/v1/2024.nlp4pi-1.29 D iversity M ed QA : A benchmark for assessing demographic biases in medical diagnosis using large language mo...

    2024 doi

  19. [27]

    Rachel Rudinger, Jason Naradowsky, Brian Leonard, and Benjamin Van Durme. 2018. https://doi.org/10.18653/v1/N18-2002 Gender bias in coreference resolution . In Proceedings of the 2018 Conference of the North A merican Chapter of the Association for Computational Linguistics: H...

    2018 doi

  20. [28]

    Smith, and Yejin Choi

    Maarten Sap, Saadia Gabriel, Lianhui Qin, Dan Jurafsky, Noah A. Smith, and Yejin Choi. 2020. https://doi.org/10.18653/v1/2020.acl-main.486 Social bias frames: Reasoning about social and power implications of language . In Proceedings of the 58th Annual Meeting of the Associati...

    2020 doi

  21. [29]

    Klaus R Scherer and Harald G Wallbott. 1994. Evidence for universality and cultural variation of differential emotion response patterning. Journal of personality and social psychology, 66(2):310

    1994

  22. [30]

    Stephen A Schullo and Burton L Alperson. 1984. Interpersonal phenomenology as a function of sexual orientation, sex, sentiment, and trait categories in long-term dyadic relationships. Journal of Personality and Social Psychology, 47(5):983

    1984

  23. [31]

    Karolina Stanczak and Isabelle Augenstein. 2021. https://arxiv.org/abs/2112.14168 A survey on gender bias in natural language processing . Preprint, arXiv:2112.14168

    2021 arXiv

  24. [32]

    Alex Tamkin, Amanda Askell, Liane Lovitt, Esin Durmus, Nicholas Joseph, Shauna Kravec, Karina Nguyen, Jared Kaplan, and Deep Ganguli. 2023. https://arxiv.org/abs/2312.03689 Evaluating and mitigating discrimination in language model decisions . Preprint, arXiv:2312.03689

    2023 arXiv

  25. [33]

    Gemma Team, Morgane Riviere, Shreya Pathak, Pier Giuseppe Sessa, Cassidy Hardin, Surya Bhupatiraju, Léonard Hussenot, Thomas Mesnard, Bobak Shahriari, Alexandre Ramé, Johan Ferret, Peter Liu, Pouya Tafti, Abe Friesen, Michelle Casbon, Sabela Ramos, Ravin Kumar, Charline Le Lan...

    2024 arXiv

  26. [34]

    Elsbeth Turcan and Kathy McKeown. 2019. https://doi.org/10.18653/v1/D19-6213 D readdit: A R eddit dataset for stress analysis in social media . In Proceedings of the Tenth International Workshop on Health Text Mining and Information Analysis (LOUHI 2019), pages 97--107, Hong K...

    2019 doi

  27. [35]

    Jesse Vig, Sebastian Gehrmann, Yonatan Belinkov, Sharon Qian, Daniel Nevo, Yaron Singer, and Stuart Shieber. 2020. Investigating gender bias in language models using causal mediation analysis. Advances in neural information processing systems, 33:12388--12401

    2020

  28. [36]

    Yixin Wan, George Pu, Jiao Sun, Aparna Garimella, Kai-Wei Chang, and Nanyun Peng. 2023. https://doi.org/10.18653/v1/2023.findings-emnlp.243 kelly is a warm person, joseph is a role model : Gender biases in LLM -generated reference letters . In Findings of the Association for C...

    2023 doi

  29. [37]

    van Buchem, Malvika Pillai, and Tina Hernandez-Boussard

    Yuqing Wang, Yun Zhao, Sara Alessandra Keller, Anne de Hond, Marieke M. van Buchem, Malvika Pillai, and Tina Hernandez-Boussard. 2024. https://arxiv.org/abs/2406.12033 Unveiling and mitigating bias in mental health analysis with large language models . Preprint, arXiv:2406.12033

    2024 arXiv

  30. [38]

    Iain Weissburg, Sathvika Anand, Sharon Levy, and Haewon Jeong. 2025. https://arxiv.org/abs/2410.14012 Llms are biased teachers: Evaluating llm bias in personalized education . Preprint, arXiv:2410.14012

    2025 arXiv

  31. [39]

    Kyra Wilson and Aylin Caliskan. 2024. Gender, race, and intersectional bias in resume screening via language model retrieval. In Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, volume 7, pages 1578--1590

    2024

  32. [40]

    Leon Yin, Davey Alba, and Leonardo Nicoletti. 2024. https://web.archive.org/web/20250301020958/https://www.bloomberg.com/graphics/2024-openai-gpt-hiring-racial-discrimination/ Openai’s gpt is a recruiter’s dream tool. tests show there’s racial bias . Accessed: 2025-04-19

    2024

  33. [41]

    Zhexin Zhang, Leqi Lei, Lindong Wu, Rui Sun, Yongkang Huang, Chong Long, Xiao Liu, Xuanyu Lei, Jie Tang, and Minlie Huang. 2024. https://doi.org/10.18653/v1/2024.acl-long.830 S afety B ench: Evaluating the safety of large language models . In Proceedings of the 62nd Annual Mee...

    2024 doi

  34. [42]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...

  35. [43]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.