Research map / Large language models / Language, safety and interpretability
BiasFreeBench: a Benchmark for Mitigating Bias in Large Language Model Responses
Xin Xu, Xunzhi He, Churan Zhi, Ruizhe Chen, Julian McAuley, Zexue He
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Apply CoT debiasing prompt | Llama-3.1-8B-Instruct · FairMT-Bench | Bias-Free Score: 76.84 -> 94.40 | Improved |
| Apply Self-Help query rewriting | Llama-3.1-8B-Instruct · BBQ (ambiguous) | Bias-Free Score: 52.41 -> 95.52 | Improved |
| Apply Self-Help query rewriting on long contexts | Llama-3.1-8B-Instruct · FairMT-Bench | Bias-Free Score: 76.84 -> 78.83 | No clear effect |
| Apply Self-Awareness bias-type warning | Qwen2.5-7B-Instruct · FairMT-Bench | Bias-Free Score: 58.83 -> 94.24 | Improved |
| Apply Self-Awareness bias-type warning | Llama-3.1-8B-Instruct · BBQ (ambiguous) | Bias-Free Score: 52.41 -> 52.55 | No clear effect |
| Train with DPO instead of SFT on StereoSet | Llama-3.1-8B-Instruct · BBQ (ambiguous) | Bias-Free Score: 52.11 -> 58.56 | Improved |
| Apply Safe RLHF | Mistral-7B-Instruct-v0.3 · BBQ (ambiguous) | Bias-Free Score: 81.24 -> 47.30 | Hurt |
| Apply Task Vector editing | Llama-3.1-8B-Instruct · BBQ (ambiguous) | Bias-Free Score: 52.41 -> 82.77 | Improved |
| Apply Task Vector editing | Llama-3.1-8B-Instruct · BoolQ | accuracy change: 85.38 -> -22.57 points | Hurt |
| Use class-balanced StereoSet training data for SFT/DPO (ablation) | Llama-3.1-8B-Instruct · BBQ (ambiguous) | Bias-Free Score: SFT 50.68 -> 52.11; DPO 55.71 -> 58.56 | Mixed |
| Use class-balanced StereoSet training data for DPO (ablation) | Llama-3.1-8B-Instruct · FairMT-Bench | Bias-Free Score: 85.88 -> 82.54 | Hurt |
| Increase Qwen2.5 model size with prompting-based debiasing (ablation) | Qwen2.5 (5 sizes) · BBQ and FairMT-Bench | Bias-Free Score | Improved |
Closest papers by meaning
- Biases Mitigation and Expressiveness Preservation in Language Models: A Comprehensive Pipeline (Student Abstract) (AAAI 2024 · similarity 0.92)
- B-score: Detecting biases in large language models using response history (ICML 2025 · similarity 0.92)
- Debiasing Algorithm through Model Adaptation (ICLR 2024 · similarity 0.91)
- Evaluating the Efficacy of Prompting Techniques for Debiasing Language Model Outputs (Student Abstract) (AAAI 2024 · similarity 0.91)
- Investigating and Mitigating Undesirable Biases in Large Language Models (AAAI 2025 · similarity 0.91)
- Any Large Language Model Can Be a Reliable Judge: Debiasing with a Reasoning-based Bias Detector (NeurIPS 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
