Metraipolis › archive › dc_4cf31dd40ef6
When and Why Do Linear Bias Probes Fail? A Geometric and Statistical Theory of Bias Detectability in Large Language Model Representations — arXiv:2609.22337
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.22337
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-09T07:08:08.926Z
- SHA-256 of the archived text
0088e681dd2432ec07eb3ce020f2f8b8021201accfcf71c8f8404fbca40e3741- Stable URL of this page
https://metraipolis.com/documento/dc_4cf31dd40ef6
How to cite
When and Why Do Linear Bias Probes Fail? A Geometric and Statistical Theory of Bias Detectability in Large Language Model Representations — arXiv:2609.22337 — arXiv (en), https://arxiv.org/abs/2609.22337 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_4cf31dd40ef6 (Metraipolis archive dc_4cf31dd40ef6, sha256 0088e681dd2432ec07eb3ce020f2f8b8021201accfcf71c8f8404fbca40e3741)
BibTeX
@misc{dc4cf31dd40ef6,
title = {When and Why Do Linear Bias Probes Fail? A Geometric and Statistical Theory of Bias Detectability in Large Language Model Representations — arXiv:2609.22337},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.22337},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_4cf31dd40ef6; sha256 0088e681dd2432ec07eb3ce020f2f8b8021201accfcf71c8f8404fbca40e3741},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.22337. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Linear probing is the standard instrument for detecting social biases in the hidden representations of large language models. Yet reported probe accuracies come almost exclusively from \emph{counterfactual} evaluations in which every input carries an explicit demographic marker. Once only a fraction $\alpha$ of inputs carries demographic information, performance degrades sharply, and a weak probe may reflect either an unbiased model or an underpowered detector. We develop a theory that resolves this ambiguity. Modeling representations as two class-conditional clusters with Mahalanobis separation $s$ on a manifold of curvature $\kap$, we prove: (i) a finite-sample generalization bound governed by the manifold's extrinsic radius with a matching $\smash{\sqrt{\dB/n}}$ minimax lower bound; (ii) an exact purity law for the maximum linear-probe AUC, strictly increasing in $\alpha$; (iii) a curvature ceiling: ambient chordal separation on a space form cannot exceed $2/\sqrt{\kap}$; and (iv) a detectability threshold below which no audit can distinguish probe output from chance. Every theorem is validated on synthetic manifolds with known ground truth and on six open-weight models $\times$ four bias dimensions, where the purity law predicts entire AUC--$\alpha$ curves from a single cross-fitted $\hat s$ measured at $\alpha=1$, with no parameters fitted to those curves. The framework turns bias auditing into a power analysis: given a target purity and effect size, it prescribes the sample budget $n(\alpha)$ for a conclusive audit.
Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-22 in cs.LG).
Categorie arXiv: Machine Learning (stat.ML) ; Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.22337 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-09T06:58:09.500Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-09-22?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).