Research map / Large language models / Language, safety and interpretability
Bias Similarity Measurement: A Black-Box Audit of Fairness Across LLMs
Hyejun Jeong, Shiqing Ma, Amir Houmansadr
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 |
|---|---|---|---|
| instruction-tune LLaMA 3.1 8B | LLaMA 3.1 8B · BBQ ambiguous | bias score s_AMB: 18.59 -> 1.38 | Improved |
| instruction-tune LLaMA 3.1 8B | LLaMA 3.1 8B · BBQ disambiguated | bias score s_DIS: 31.37 -> 4.78 | Improved |
| instruction-tune LLaMA 3.2 3B | LLaMA 3.2 3B · BBQ ambiguous | bias score s_AMB: 11.95 -> 15.71 | Hurt |
| instruction-tune Gemma 3 4B | Gemma 3 4B · BBQ ambiguous | bias score s_AMB: -3.89 -> 5.83 | Hurt |
| instruction-tune Gemma 2 27B | Gemma 2 27B · BBQ ambiguous | bias score s_AMB: 6.95 -> 0.51 | Improved |
| instruction-tune LLaMA 2 7B | LLaMA 2 7B · BBQ ambiguous | bias score s_AMB: 5.45 -> 4.30 | No clear effect |
| instruction-tune models (accuracy) | LLMs · BBQ disambiguated | accuracy | Improved |
| switch from BBQ (abstention allowed) to UnQover (forced choice) | instruction-tuned LLMs · BBQ vs UnQover | stereotypical preference | Hurt |
| instruction tuning effect on internal representations | base vs tuned models · CKA | diagonal CKA: 0.941 to 0.991 | No clear effect |
Closest papers by meaning
- Prompting Fairness: Integrating Causality to Debias Large Language Models (ICLR 2025 · similarity 0.90)
- StereoTales: A Multilingual Framework for Open-Ended Stereotype Discovery in LLMs (NeurIPS 2026 · similarity 0.88)
- Distributive Fairness in Large Language Models: Evaluating Alignment with Human Values (NeurIPS 2025 · similarity 0.88)
- All Should Be Equal in the Eyes of LMs: Counterfactually Aware Fair Text Generation (AAAI 2024 · similarity 0.88)
- Biases Mitigation and Expressiveness Preservation in Language Models: A Comprehensive Pipeline (Student Abstract) (AAAI 2024 · similarity 0.88)
- FairMT-Bench: Benchmarking Fairness for Multi-turn Dialogue in Conversational LLMs (ICLR 2025 · similarity 0.88)
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.
