Metraipolis › archive › dc_62aba07334a1

Interpretability Is Not Model-Specific — arXiv:2611.12675

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.12675
License (as archived)
CC0-1.0
Archived
2026-11-10T18:41:55.738Z
SHA-256 of the archived text
04a8b826265a54a77e655abf928d7f9845c672c21a5717372ccf35f9faf72f0a
Stable URL of this page
https://metraipolis.com/documento/dc_62aba07334a1

How to cite

Interpretability Is Not Model-Specific — arXiv:2611.12675 — arXiv (en), https://arxiv.org/abs/2611.12675 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_62aba07334a1 (Metraipolis archive dc_62aba07334a1, sha256 04a8b826265a54a77e655abf928d7f9845c672c21a5717372ccf35f9faf72f0a)
BibTeX
@misc{dc62aba07334a1,
  title = {Interpretability Is Not Model-Specific — arXiv:2611.12675},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.12675},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_62aba07334a1; sha256 04a8b826265a54a77e655abf928d7f9845c672c21a5717372ccf35f9faf72f0a},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.12675. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Interpretability tools are usually built for one model at a time: a refusal direction, a truthfulness probe, or a steering vector is found with labeled data inside the model it explains. We ask whether such tools can be carried to other models through a map between residual streams fit only on unlabeled prompts. We transfer a refusal direction, tested causally by addition and ablation, and two read-out tools, a correctness direction and a sentiment probe, across 110 pairs of models from five families and 1B to 72B parameters and between base and instruction-tuned models, with controls for specificity and five seeds. How much transfers depends on what the map was fit on. A map fit on ordinary questions recovers 54\% of the target's own refusal steering and 16\% to 77\% of its correctness gain; adding unlabeled prompts from the region where the tool operates raises these to 102\% and 88\% to 98\%, while control directions recover at most 1\%. We prove that a linear map behaves this way whenever two models correspond linearly on the data it covers, and a label-free coverage score derived from this result tracks recovery across more than 16,000 runs and explains earlier reports that maps between models fail across domains. Carried read-out tools outperform anything the target's unlabeled representations yield. Interpretability results, once found, can be reused across models at the cost of a few unlabeled prompts.

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/2611.12675 — 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 2026-11-10T18:37:37.554Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).