Metraipolis › archive › dc_3de90401bc4f

Deep Noir: Autonomous Steering Discovery via Architectural Chronometry in Transformer Models — arXiv:2609.20722

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.20722
License (as archived)
CC0-1.0
Archived
2026-10-10T03:22:19.067Z
SHA-256 of the archived text
e222903b6ee56f7aead9cea018a7e9be9d4dd014d001c3b059cadc0bb329bc1e
Stable URL of this page
https://metraipolis.com/documento/dc_3de90401bc4f

How to cite

Deep Noir: Autonomous Steering Discovery via Architectural Chronometry in Transformer Models — arXiv:2609.20722 — arXiv (en), https://arxiv.org/abs/2609.20722 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_3de90401bc4f (Metraipolis archive dc_3de90401bc4f, sha256 e222903b6ee56f7aead9cea018a7e9be9d4dd014d001c3b059cadc0bb329bc1e)
BibTeX
@misc{dc3de90401bc4f,
  title = {Deep Noir: Autonomous Steering Discovery via Architectural Chronometry in Transformer Models — arXiv:2609.20722},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.20722},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_3de90401bc4f; sha256 e222903b6ee56f7aead9cea018a7e9be9d4dd014d001c3b059cadc0bb329bc1e},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.20722. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Activation steering modifies LLM behavior at inference time, but identifying where and how strongly to steer remains manual. We introduce Deep Noir, a framework that uses Logit Lens convergence and causal head-level attribution to autonomously discover optimal steering parameters. Across three scales (1B x 3, 2-3B x 2, and 7-9B x 4), our engine achieves 16.7 percentage-point improvement on spam at 1B (standard deviation 4.7; 39 runs), with gains increasing to 21 to 42 percentage points at 7-9B across four architectures. On SST-2 sentiment, it achieves a 13.1 percentage-point improvement with zero code changes. Mechanistic grounding enables automated discovery of intervention points that generalize across tasks and architectures. On sentiment, RepE without head masking fails to improve over baseline, while Deep Noir improves all models (p less than 0.01). We further show that steering creates a predictable prompt-injection attack surface whose vulnerability increases monotonically with steering magnitude. This finding is relevant to agent systems deploying steered classifiers.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-18 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.20722 — 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 2026-10-10T03:04:13.758Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-18?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).