Metraipolis › archive › dc_bcda4517f8b1
The Geometry of Ignorance: Detecting Suppressed Knowledge in LLMs through Activation Fragmentation — arXiv:2611.05272
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.05272
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-04T12:28:46.509Z
- SHA-256 of the archived text
633ac581a29cf96f30db42bcb8d317fd5a7073da893f2ff9345cc788f909c5f2- Stable URL of this page
https://metraipolis.com/documento/dc_bcda4517f8b1
How to cite
The Geometry of Ignorance: Detecting Suppressed Knowledge in LLMs through Activation Fragmentation — arXiv:2611.05272 — arXiv (en), https://arxiv.org/abs/2611.05272 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_bcda4517f8b1 (Metraipolis archive dc_bcda4517f8b1, sha256 633ac581a29cf96f30db42bcb8d317fd5a7073da893f2ff9345cc788f909c5f2)
BibTeX
@misc{dcbcda4517f8b1,
title = {The Geometry of Ignorance: Detecting Suppressed Knowledge in LLMs through Activation Fragmentation — arXiv:2611.05272},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.05272},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_bcda4517f8b1; sha256 633ac581a29cf96f30db42bcb8d317fd5a7073da893f2ff9345cc788f909c5f2},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.05272. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Evaluating whether a large language model (LLM) genuinely knows a concept remains a fundamental problem, since behavioral metrics can be fooled by suppression: a model that refuses to answer may still harbor recoverable knowledge. The framework measures the geometric coherence of \textit{anon-vectors}, the differences in hidden states between a concept-specific query and its anonymized paraphrase, and shows that genuine knowledge forms a tight, low-entropy cluster while ignorance forces geometric fragmentation. Across ten instruction-tuned LLMs from three model families, all fragmentation metrics reliably separate known from unknown concepts, and post-unlearning, fragmentation geometry correlates with adversarial recovery across four attack types and five unlearning algorithms, though the direction of that relationship is shown to depend on attack methodology. Furthermore, a layer-resolved analysis reveals that different adversarial mechanisms exploit this geometry at varying depths, with high attack-specific correlations that are invisible to behavioral evaluation. Together, these results constitute a single-phase, attack-free proxy for unlearning robustness that captures aspects of parametric knowledge inaccessible to output-level metrics. A reference implementation is available at this https URL
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/2611.05272 — 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 2026-11-04T12:26:47.198Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).