Beyond Bias Scores: Answerability Confounds in Activation Steering with Gradient-Constrained Direction Design (GCD)
Bias scores in language-model evaluations often conflate what a model prefers with whether it answers at all. We identify this confound in \textsc{BBQ}, whose ambiguous score factorizes into answerability and signed preference. Across 11 categories, 28,}824 items, and five models, contrastive steering lowers the signed ambiguous score in 11/11 categories but lowers the signed disambiguated preference score in only 8/11. Across 99 configurations, the answerability channel exceeds half of the observed score change in 17\% and accounts for the entire change in 11%. In a six-category disambiguated evaluation, 20.8\% of 1,500 items flip to the unknown option, where abstention is incorrect; \te...
| Authors | Weitong Li, Donglin Song, Junkai Lin et al. |
|---|---|
| Primary category | cs.AI |
| Published | 13 October 2026 |
| Last updated at source | 13 October 2026 |
Silytics shows public metadata from arXiv and links to the original. Information may be incomplete or out of date. Last refreshed 15 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.