LLMs show up to 60.58% social bias in generated code; a new Fairness Monitor Agent cuts bias by 65.1% and raises functional correctness from 75.80% to 83.97%.
hub Mixed citations
BBQ: A Hand-Built Bias Benchmark for Question Answering
Mixed citation behavior. Most common role is background (67%).
29
Pith papers citing it
Background
67% of classified citations
abstract
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.
hub tools
citation-role summary
background 4
dataset 2
citation-polarity summary
representative citing papers
BiasedTales-ML provides a parallel multilingual corpus of LLM-generated children's stories that reveals substantial cross-lingual differences in narrative attributes not captured by English-centric analyses.
PRIMETIME generator reveals that LLM datetime parsing and arithmetic primitives are individually unreliable but fully learnable via fine-tuning, enabling frontier-level accuracy on event planning with small LoRA models.
AgentFairBench is a multi-domain benchmark for demographic disparity in LLM agent actions, with a pilot showing no significant effect for Claude Haiku 4.5 after arity-matched noise correction.
Configuration choices alone flip pairwise safety verdicts on every tested alignment benchmark, isolated via a finite-envelope proposition linking disagreement rate to strict ordering reversal.
Toxicity benchmarks for LLMs produce inconsistent results when task type, input domain, or model changes, revealing intrinsic evaluation biases.
Parcae stabilizes looped LLMs via spectral norm constraints on injection parameters, enabling power-law scaling for training FLOPs and saturating exponential scaling at test time that improves quality over fixed-depth baselines under fixed parameter budgets.
Multi-agent systems amplify minor stochastic biases into systemic polarization via echo-chamber effects in structured workflows, even with neutral agents.
Kimi K2.5 matches closed models on dual-use tasks but refuses fewer CBRNE requests and shows some sabotage and self-replication tendencies.
Language models refuse 75.4% of requests to evade defeated rules and do so even after recognizing reasons that undermine the rule's legitimacy.
OutSafe-Bench supplies the first large-scale four-modality safety dataset and evaluation framework that exposes persistent unsafe outputs in nine leading multimodal LLMs.
Empirical analysis shows scaling inference compute via strategies like tree search can be more efficient than scaling model parameters, with 7B models plus novel search outperforming 34B models.
Gemini Ultra reaches human-expert performance on MMLU for the first time and sets new state-of-the-art results on 30 of 32 benchmarks, including all 20 multimodal ones tested.
Language models show good calibration when asked to estimate the probability that their own answers are correct, with performance improving as models get larger.
Emergent abilities are capabilities present in large language models but absent in smaller ones and cannot be predicted by extrapolating smaller model performance.
RLHF alignment training on language models boosts NLP performance, supports skill specialization, enables weekly online updates with fresh human data, and shows a linear relation between RL reward and sqrt(KL divergence from initialization.
A multi-role red-teaming framework with attacker, target, and jury LLMs measures faithfulness in English and Arabic, finding false-premise prompts and length limits change unfaithfulness rates.
Ghostwriter attack injects fabricated evidence to steer LLM viewpoints, with experiments showing high success on commercial models and partial mitigation on guarded ones.
LLMs generate narratives containing persistent stereotypes, erasure, and one-dimensional portrayals of Global Majority national identities, with minoritized groups overrepresented in subordinated roles by more than fifty times compared to dominant portrayals.
LLMs are more accurate when answers match stereotypes in clear contexts, especially for race-gender combinations, and no tested model shows consistent fairness or reliability across intersectional groups.
OpenAI releases two open-weight reasoning models, gpt-oss-120b and gpt-oss-20b, trained via distillation and RL with claimed strong results on math, coding, and safety benchmarks.
Mixtral 8x7B is a sparse MoE LLM activating 2 of 8 experts per layer that matches or exceeds Llama 2 70B and GPT-3.5 on benchmarks while using only 13B active parameters.
PaLM 2 reports state-of-the-art results on language, reasoning, and multilingual tasks with improved efficiency over PaLM.
OpenAI reports that chain-of-thought reasoning in o1 models enables deliberative alignment, yielding state-of-the-art results on selected safety benchmarks for illicit advice, stereotypes, and jailbreaks.
citing papers explorer
-
Social Bias in LLM-Generated Code: Benchmark and Mitigation
LLMs show up to 60.58% social bias in generated code; a new Fairness Monitor Agent cuts bias by 65.1% and raises functional correctness from 75.80% to 83.97%.
-
BIASEDTALES-ML: A Multilingual Dataset for Analyzing Narrative Attribute Distributions in LLM-Generated Stories
BiasedTales-ML provides a parallel multilingual corpus of LLM-generated children's stories that reveals substantial cross-lingual differences in narrative attributes not captured by English-centric analyses.
-
PRIMETIME : Limits of LLMs in Temporal Primitives
PRIMETIME generator reveals that LLM datetime parsing and arithmetic primitives are individually unreliable but fully learnable via fine-tuning, enabling frontier-level accuracy on event planning with small LoRA models.
-
AgentFairBench: Do LLM Agents Discriminate When They Act?
AgentFairBench is a multi-domain benchmark for demographic disparity in LLM agent actions, with a pilot showing no significant effect for Claude Haiku 4.5 after arity-matched noise correction.
-
SafetyRepro: Configuration-Conditional Rank Instability on Alignment Benchmarks
Configuration choices alone flip pairwise safety verdicts on every tested alignment benchmark, isolated via a finite-envelope proposition linking disagreement rate to strict ordering reversal.
-
Navigating the Sea of LLM Evaluation: Investigating Bias in Toxicity Benchmarks
Toxicity benchmarks for LLMs produce inconsistent results when task type, input domain, or model changes, revealing intrinsic evaluation biases.
-
Parcae: Scaling Laws For Stable Looped Language Models
Parcae stabilizes looped LLMs via spectral norm constraints on injection parameters, enabling power-law scaling for training FLOPs and saturating exponential scaling at test time that improves quality over fixed-depth baselines under fixed parameter budgets.
-
Aligned Agents, Biased Swarm: Measuring Bias Amplification in Multi-Agent Systems
Multi-agent systems amplify minor stochastic biases into systemic polarization via echo-chamber effects in structured workflows, even with neutral agents.
-
An Independent Safety Evaluation of Kimi K2.5
Kimi K2.5 matches closed models on dual-use tasks but refuses fewer CBRNE requests and shows some sabotage and self-replication tendencies.
-
Blind Refusal: Language Models Refuse to Help Users Evade Unjust, Absurd, and Illegitimate Rules
Language models refuse 75.4% of requests to evade defeated rules and do so even after recognizing reasons that undermine the rule's legitimacy.
-
OutSafe-Bench: A Benchmark for Multimodal Offensive Content Detection in Large Language Models
OutSafe-Bench supplies the first large-scale four-modality safety dataset and evaluation framework that exposes persistent unsafe outputs in nine leading multimodal LLMs.
-
Inference Scaling Laws: An Empirical Analysis of Compute-Optimal Inference for Problem-Solving with Language Models
Empirical analysis shows scaling inference compute via strategies like tree search can be more efficient than scaling model parameters, with 7B models plus novel search outperforming 34B models.
-
Gemini: A Family of Highly Capable Multimodal Models
Gemini Ultra reaches human-expert performance on MMLU for the first time and sets new state-of-the-art results on 30 of 32 benchmarks, including all 20 multimodal ones tested.
-
Language Models (Mostly) Know What They Know
Language models show good calibration when asked to estimate the probability that their own answers are correct, with performance improving as models get larger.
-
Emergent Abilities of Large Language Models
Emergent abilities are capabilities present in large language models but absent in smaller ones and cannot be predicted by extrapolating smaller model performance.
-
Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
RLHF alignment training on language models boosts NLP performance, supports skill specialization, enables weekly online updates with fresh human data, and shows a linear relation between RL reward and sqrt(KL divergence from initialization.
-
A Red Teaming Framework for Large Language Models: A Case Study on Faithfulness Evaluation
A multi-role red-teaming framework with attacker, target, and jury LLMs measures faithfulness in English and Arabic, finding false-premise prompts and length limits change unfaithfulness rates.
-
Steering LLM Viewpoints through Fabricated Evidence Injection
Ghostwriter attack injects fabricated evidence to steer LLM viewpoints, with experiments showing high success on commercial models and partial mitigation on guarded ones.
-
Representational Harms in LLM-Generated Narratives Against Global Majority Nationalities
LLMs generate narratives containing persistent stereotypes, erasure, and one-dimensional portrayals of Global Majority national identities, with minoritized groups overrepresented in subordinated roles by more than fifty times compared to dominant portrayals.
-
Intersectional Fairness in Large Language Models
LLMs are more accurate when answers match stereotypes in clear contexts, especially for race-gender combinations, and no tested model shows consistent fairness or reliability across intersectional groups.
-
gpt-oss-120b & gpt-oss-20b Model Card
OpenAI releases two open-weight reasoning models, gpt-oss-120b and gpt-oss-20b, trained via distillation and RL with claimed strong results on math, coding, and safety benchmarks.
-
Mixtral of Experts
Mixtral 8x7B is a sparse MoE LLM activating 2 of 8 experts per layer that matches or exceeds Llama 2 70B and GPT-3.5 on benchmarks while using only 13B active parameters.
-
PaLM 2 Technical Report
PaLM 2 reports state-of-the-art results on language, reasoning, and multilingual tasks with improved efficiency over PaLM.
-
OpenAI o1 System Card
OpenAI reports that chain-of-thought reasoning in o1 models enables deliberative alignment, yielding state-of-the-art results on selected safety benchmarks for illicit advice, stereotypes, and jailbreaks.
-
OpenAI GPT-5 System Card
GPT-5 is a unified model system that routes queries between fast and deep reasoning paths and reports gains in real-world usefulness, reduced hallucinations, and safety features over prior versions.
-
Fairness in Multi-Agent Systems for Software Engineering: An SDLC-Oriented Rapid Review
A rapid review of fairness in LLM-enabled multi-agent systems for the software development lifecycle concludes that the field lacks standardized evaluations, broad coverage, and effective governance, leaving it unprepared for deployable fair systems.
-
The nextAI Solution to the NeurIPS 2023 LLM Efficiency Challenge
A competition entry achieved efficient fine-tuning of LLaMa2 70B on one GPU in 24 hours with competitive QA benchmark performance.
- RepSelect: Robust LLM Unlearning via Representation Selectivity
- In-Situ Behavioral Evaluation for LLM Fairness, Not Standardized-Test Scores