Metraipolis › archive › dc_4fc7a69be868
Sensitivity, Causality, and Repair Dissociate: A Layer-Wise Analysis of Perturbation Robustness and Its Scaling — arXiv:2608.03842
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.03842
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-31T12:31:11.653Z
- SHA-256 of the archived text
8f1b4b9f5721abede5b95d37ea5c91b05f937457a85be096ca1727d7b2b2e604- Stable URL of this page
https://metraipolis.com/documento/dc_4fc7a69be868
How to cite
Sensitivity, Causality, and Repair Dissociate: A Layer-Wise Analysis of Perturbation Robustness and Its Scaling — arXiv:2608.03842 — arXiv (en), https://arxiv.org/abs/2608.03842 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_4fc7a69be868 (Metraipolis archive dc_4fc7a69be868, sha256 8f1b4b9f5721abede5b95d37ea5c91b05f937457a85be096ca1727d7b2b2e604)
BibTeX
@misc{dc4fc7a69be868,
title = {Sensitivity, Causality, and Repair Dissociate: A Layer-Wise Analysis of Perturbation Robustness and Its Scaling — arXiv:2608.03842},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.03842},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_4fc7a69be868; sha256 8f1b4b9f5721abede5b95d37ea5c91b05f937457a85be096ca1727d7b2b2e604},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.03842. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
When a language model fails on surface-perturbed input (typos, OCR noise, homophones), "which layer is responsible" has three natural operationalizations: where representations diverge most (sensitivity), where restoring clean activations recovers the prediction (causality), and where a small adapter can repair the damage (compensatory capacity) - and we show these three layer maps dissociate. Across a five-model panel we identify two propagation regimes - spike-and-suppress (Phi-3.5, Gemma-2-9B) and late-accumulation (Llama-3, Mistral, Qwen2.5-7B) - and on the two models meeting an 80% identity-patch gate, sensitivity and causality are anti-correlated (rho = -0.72 to -0.88). Within-family scaling on Qwen2.5 (1.5B to 14B) shows the late-accumulation signature strengthening monotonically with scale, corroborated on a second family. We propose cascade disruption as the mechanism behind the dissociation: adapters placed at causally implicated early layers break intact downstream computation, making diagnostic-flagged sites the worst adapter placements. A fixed-harness layer sweep across four models (3.8-8B) confirms the core prediction on chain-of-thought GSM8K - the flagged sites are the most damaging adapter windows on every adjudicable model - and is sign-consistent but strongly attenuated on a multiple-choice control, consistent with damage that compounds with generation length. The sweep yields practical guidance: a training-free LRD pre-screen and a default-deepest placement rule, though absolute gains over no-adapter baselines remain small. Finally, apparent gains from a representation-stability loss reverse under an adequate generation budget - truncated chain-of-thought had been scored as empty - a methodological warning for any intervention evaluated on chain-of-thought tasks. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-05 in cs.CL). Categorie arXiv: Computation and Language (cs.CL) ; Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2608.03842 — 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 2026-10-31T12:17:37.782Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-08-05?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).