mlsanityData to 7 Oct 2026

BBQ: A Hand-Built Bias Benchmark for Question Answering

arXiv 2110.08193 ↗First posted 15 Oct 2021cs.CLcited by 683

Not on the map. Why?

Cited by 60 in LLM alignment / RLHF / preference optimisation, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

It is well documented that NLP models learn social biases, but little work has been done on how these biases manifest in model outputs for applied tasks like question answering (QA). We introduce the Bias Benchmark for QA (BBQ), a dataset of question sets constructed by the authors that highlight attested social biases against people belonging to protected classes along nine social dimensions relevant for U.S. English-speaking contexts. Our task evaluates model responses at two levels: (i) given an under-informative context, we test how strongly responses reflect social biases, and (ii) given an adequately informative context, we test whether the model's biases override a correct answer choice. We find that models often rely on stereotypes when the context is under-informative, meaning the model's outputs consistently reproduce harmful biases in this setting. Though models are more accurate when the context provides an informative answer, they still rely on stereotypes and average up to 3.4 percentage points higher accuracy when the correct answer aligns with a social bias than when it conflicts, with this difference widening to over 5 points on examples targeting gender for most models tested.

Key facts. Cited by 683 papers in the map (data through 7 Oct 2026). Builds on 23 of its 41 references that are in the map. First posted on arXiv 15 Oct 2021.

Builds on

23 of 41 references.

References · 23

  1. 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,266
  2. 2018Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challengecited by 4,710
  3. 2020Language (Technology) is Power: A Critical Survey of "Bias" in NLPcited by 951
  4. 2016Semantics derived automatically from language corpora contain human-like biasescited by 894
  5. 2017RACE: Large-scale ReAding Comprehension Dataset From Examinationscited by 851
  6. 2018Gender Bias in Coreference Resolution: Evaluation and Debiasing Methodscited by 657
  7. 2019The Woman Worked as a Babysitter: On Biases in Language Generationcited by 540
  8. 2020UnifiedQA: Crossing Format Boundaries With a Single QA Systemcited by 475
  9. 2018Gender Bias in Coreference Resolutioncited by 438
  10. 2019On Measuring Social Biases in Sentence Encoderscited by 397
  11. 2019Social Bias Frames: Reasoning about Social and Power Implications of Languagecited by 377
  12. 2020Learning from the Worst: Dynamically Generated Datasets to Improve Online Hate Detectioncited by 217
11 more
  1. 2020HateCheck: Functional Tests for Hate Speech Detection Modelscited by 215
  2. 2019Identifying and Reducing Gender Bias in Word-Level Language Modelscited by 214
  3. 2018Gender Bias in Neural Natural Language Processingcited by 210
  4. 2019Racial Bias in Hate Speech and Abusive Language Detection Datasetscited by 208
  5. 2020UnQovering Stereotyping Biases via Underspecified Questionscited by 124
  6. 2019On Measuring and Mitigating Biased Inferences of Word Embeddingscited by 112
  7. 2019Toward Gender-Inclusive Coreference Resolutioncited by 84
  8. 2019What's in a Name? Reducing Bias in Bios without Access to Protected Attributescited by 51
  9. 2021Evaluating Gender Bias in Natural Language Inferencecited by 27
  10. 2021Intersectional Bias in Causal Language Modelscited by 24
  11. 2019Resolving Gendered Ambiguous Pronouns with BERTcited by 16

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 15; 2023: 74; 2024: 171; 2025: 211; 2026: 212.

683 papers.

Sort:

By area:

  1. 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,634
  2. 2022Emergent Abilities of Large Language Modelscited by 2,483
  3. 2024Mixtral of Expertscited by 1,785
  4. 2025gpt-oss-120b & gpt-oss-20b Model Cardcited by 1,258
  5. 2023A Survey on Evaluation of Large Language Modelscited by 1,227
  6. 2024OpenAI o1 System Cardcited by 1,220
  7. 2023PaLM 2 Technical Reportcited by 1,154
  8. 2022When Not to Trust Language Models: Investigating Effectiveness of Parametric and Non-Parametric Memoriescited by 1,103
  9. 2023Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Promptingcited by 1,081
  10. 2023BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Datasetcited by 828
  11. 2022Discovering Language Model Behaviors with Model-Written Evaluationscited by 805
  12. 2023A Comprehensive Overview of Large Language Modelscited by 639
38 more
  1. 2023Bias and Fairness in Large Language Models: A Surveycited by 626
  2. 2022Improving alignment of dialogue agents via targeted human judgementscited by 527
  3. 2023Principle-Driven Self-Alignment of Language Models from Scratch with Minimal Human Supervisioncited by 318
  4. 2024Deliberative Alignment: Reasoning Enables Safer Language Modelscited by 292
  5. 2022Prompting GPT-3 To Be Reliablecited by 258
  6. 2023Large Language Model Alignment: A Surveycited by 241
  7. 2024SORRY-Bench: Systematically Evaluating Large Language Model Safety Refusalcited by 224
  8. 2023Red-Teaming Large Language Models using Chain of Utterances for Safety-Alignmentcited by 213
  9. 2023Language Model Tokenizers Introduce Unfairness Between Languagescited by 207
  10. 2023Inverse Scaling: When Bigger Isn't Bettercited by 205
  11. 2025Why Language Models Hallucinatecited by 205
  12. 2023Marked Personas: Using Natural Language Prompts to Measure Stereotypes in Language Modelscited by 204
  13. 2023FLASK: Fine-grained Language Model Evaluation based on Alignment Skill Setscited by 198
  14. 2022On Second Thought, Let's Not Think Step by Step! Bias and Toxicity in Zero-Shot Reasoningcited by 191
  15. 2023SafetyBench: Evaluating the Safety of Large Language Modelscited by 189
  16. 2024Lessons from the Trenches on Reproducible Evaluation of Language Modelscited by 172
  17. 2023The Capacity for Moral Self-Correction in Large Language Modelscited by 171
  18. 2023Defending Large Language Models Against Jailbreaking Attacks Through Goal Prioritizationcited by 156
  19. 2025AbstentionBench: Reasoning LLMs Fail on Unanswerable Questionscited by 129
  20. 2023Evaluating and Mitigating Discrimination in Language Model Decisionscited by 117
  21. 2024Twin-Merging: Dynamic Integration of Modular Expertise in Model Mergingcited by 114
  22. 2024BetterBench: Assessing AI Benchmarks, Uncovering Issues, and Establishing Best Practicescited by 112
  23. 2023Mechanistically analyzing the effects of fine-tuning on procedurally defined taskscited by 110
  24. 2023Generative Language Models Exhibit Social Identity Biasescited by 107
  25. 2024Language models scale reliably with over-training and on downstream taskscited by 105
  26. 2022State-of-the-art generalisation research in NLP: A taxonomy and reviewcited by 97
  27. 2024OffsetBias: Leveraging Debiased Data for Tuning Evaluatorscited by 94
  28. 2023Clarify When Necessary: Resolving Ambiguity Through Interaction with LMscited by 93
  29. 2023Evaluating the Social Impact of Generative AI Systems in Systems and Societycited by 91
  30. 2023A Survey on Fairness in Large Language Modelscited by 91
  31. 2024Large Language Models are Geographically Biasedcited by 91
  32. 2024KMMLU: Measuring Massive Multitask Language Understanding in Koreancited by 91
  33. 2023Red teaming ChatGPT via Jailbreaking: Bias, Robustness, Reliability and Toxicitycited by 86
  34. 2024Aligning to Thousands of Preferences via System Message Generalizationcited by 84
  35. 2024InferAligner: Inference-Time Alignment for Harmlessness through Cross-Model Guidancecited by 78
  36. 2025LLM Generated Persona is a Promise with a Catchcited by 77
  37. 2023Use of LLMs for Illicit Purposes: Threats, Prevention Measures, and Vulnerabilitiescited by 76
  38. 2024SafetyPrompts: a Systematic Review of Open Datasets for Evaluating and Improving Large Language Model Safetycited by 75

50 most cited of 683 All 683 (JSON)