Reference-Based Bias Detection in LLMs via Relative Representations of Hidden States
Researchers propose auditing LLM bias via relative hidden-state representations, detecting bias increases with 3-50x less compute than output-level benchmarks.
The paper introduces a reference-based bias auditing method that compares hidden-state representations across model variants, such as before and after fine-tuning, by encoding sentences relative to a fixed anchor set. The resulting Representational Bias Shift (Delta-B) correlates with output-level bias change in 15 of 18 tested settings, reaching |r| = 0.84 under full fine-tuning across WildGuardMix, DecodingTrust, and ToxiGen benchmarks. Thresholding Delta-B detects checkpoints whose bias increased with ROC AUC between 0.65 and 0.99 and beats a SEAT-based baseline, while auditing a model in about three minutes with 3-50x less compute.
- Method encodes sentences via similarities to anchor sentences, creating comparable representations across variants.
- Delta-B correlates with output-level bias change in 15 of 18 tested settings.
- Detects checkpoints with increased bias at ROC AUC 0.65-0.99, outperforming SEAT baseline.
- Audits a model in about three minutes using 3-50x less compute.
From the original article
1 min read at huggingface.co
Existing bias auditing methods typically rely on model outputs, requiring costly benchmarks or judge models and potentially missing internal shifts that never appear in generated text. We propose a reference-based method that audits bias in hidden-state representations across related model variants, for example before and after fine-tuning.
By Marek Jeliński, Jan Dubiński, Maciej Chrabaszcz · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.