Metraipolis › archive › dc_43da07a5db11
Reducing Logprob Without Causal Verification: Diagnosing SAE-Based PII Suppression in LLMs — arXiv:2611.04782
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.04782
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-04T21:38:05.610Z
- SHA-256 of the archived text
5b0328c56a27bf6bfc714fdf875fd6eb918b22c46b44dda315e0f306355ef437- Stable URL of this page
https://metraipolis.com/documento/dc_43da07a5db11
How to cite
Reducing Logprob Without Causal Verification: Diagnosing SAE-Based PII Suppression in LLMs — arXiv:2611.04782 — arXiv (en), https://arxiv.org/abs/2611.04782 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_43da07a5db11 (Metraipolis archive dc_43da07a5db11, sha256 5b0328c56a27bf6bfc714fdf875fd6eb918b22c46b44dda315e0f306355ef437)
BibTeX
@misc{dc43da07a5db11,
title = {Reducing Logprob Without Causal Verification: Diagnosing SAE-Based PII Suppression in LLMs — arXiv:2611.04782},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.04782},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_43da07a5db11; sha256 5b0328c56a27bf6bfc714fdf875fd6eb918b22c46b44dda315e0f306355ef437},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.04782. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
SAE-based feature-level interventions to suppress PII leakage in LLMs are primarily evaluated by log-probability reduction. We show that this criterion cannot distinguish feature specific suppression from a magnitude artifact: a direction-agnostic effect of sufficiently large activation perturbations. Across 36 (model, layer, PII type) cases, we compare selected SAE feature interventions with coherencematched random direction controls using logprob and rank-aware metrics. While 17 cases (47%) appear successful under logprob-only evaluation, only 1 case (2.8%) passes our multimetric framework. The rest reveal two failure modes: random directions outperform selected features, and feature selection collapse across PII types. These results indicate that existing evaluations cannot verify whether interventions target causal pathways of PII leakage. A 9,000 run causal mediation feasibility study corroborates this: the verifier separates candidates reliably, but correlation-selected candidates rarely pass it. We therefore propose a new research mission for NLP: causally grounded PII suppression through circuit-level localization and causal mediation analysis. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2611.04782 — 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 2026-11-04T21:35:49.563Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).