Metraipolis › archive › dc_cb54b75277fb

Encoded but Disconnected: Decomposing Vision-Language Model Failures under a Patching Null — arXiv:2610.00024

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.00024
License (as archived)
CC0-1.0
Archived
2026-10-06T13:56:34.271Z
SHA-256 of the archived text
71421c8e6f4a7bc7de64208b1959bbeea79439360d2d44b50e0932f936435eea
Stable URL of this page
https://metraipolis.com/documento/dc_cb54b75277fb

How to cite

Encoded but Disconnected: Decomposing Vision-Language Model Failures under a Patching Null — arXiv:2610.00024 — arXiv (en), https://arxiv.org/abs/2610.00024 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_cb54b75277fb (Metraipolis archive dc_cb54b75277fb, sha256 71421c8e6f4a7bc7de64208b1959bbeea79439360d2d44b50e0932f936435eea)
BibTeX
@misc{dccb54b75277fb,
  title = {Encoded but Disconnected: Decomposing Vision-Language Model Failures under a Patching Null — arXiv:2610.00024},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.00024},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_cb54b75277fb; sha256 71421c8e6f4a7bc7de64208b1959bbeea79439360d2d44b50e0932f936435eea},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.00024. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Across three vision-language model architectures (LLaVA-1.5-7B, Qwen2.5-VL-7B, InternVL3-8B), we report a universal negative finding for mid-layer interpretability. On POPE -- the benchmark common to all three -- the mid layers encode the ground-truth answer in 68-91% of errors, yet this signal is not causally active for the final prediction: residual-stream patching yields 0% non-trivial flip at the layer level on all three architectures, and on two of three at the per-head level (Qwen: 0/12,600 patched forwards). The lone exception, InternVL3 layer-20 head-2, is a non-vocab, self-attending head whose effect is localized to that specific head (p < 1e-4). Despite the null, the errors separate operationally into three failure modes -- Perception Failure, Encoded-but-Disconnected, Prior-Override -- learnable above 60% on all three architectures, and the architecture's prior direction predicts which of two interventions elicits a category-specific response. We report these mitigation effects under oracle labels as evidence the categories are mechanistically real, not as a deployable method.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-10-02 in cs.CL).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Computation and Language (cs.CL); Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2610.00024 — 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 2026-10-06T12:15:34.776Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-10-02?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).