AI Paper Atlas

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

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Bias measurement and mitigation · 6 citations (OpenAlex)

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 changedWhereMeasuredEffect
instruction-tune LLaMA 3.1 8BLLaMA 3.1 8B · BBQ ambiguousbias score s_AMB: 18.59 -> 1.38Improved
instruction-tune LLaMA 3.1 8BLLaMA 3.1 8B · BBQ disambiguatedbias score s_DIS: 31.37 -> 4.78Improved
instruction-tune LLaMA 3.2 3BLLaMA 3.2 3B · BBQ ambiguousbias score s_AMB: 11.95 -> 15.71Hurt
instruction-tune Gemma 3 4BGemma 3 4B · BBQ ambiguousbias score s_AMB: -3.89 -> 5.83Hurt
instruction-tune Gemma 2 27BGemma 2 27B · BBQ ambiguousbias score s_AMB: 6.95 -> 0.51Improved
instruction-tune LLaMA 2 7BLLaMA 2 7B · BBQ ambiguousbias score s_AMB: 5.45 -> 4.30No clear effect
instruction-tune models (accuracy)LLMs · BBQ disambiguatedaccuracyImproved
switch from BBQ (abstention allowed) to UnQover (forced choice)instruction-tuned LLMs · BBQ vs UnQoverstereotypical preferenceHurt
instruction tuning effect on internal representationsbase vs tuned models · CKAdiagonal CKA: 0.941 to 0.991No clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy