Metraipolis › archive › dc_9334007d5a67
Compositional Multilingual and Behavioral Attribute Steering — arXiv:2609.08410
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.08410
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-16T12:48:53.919Z
- SHA-256 of the archived text
1abfe5e9902febd7c4b5c680c99db146e67ce4c0fac7a60599f81de535c28bb5- Stable URL of this page
https://metraipolis.com/documento/dc_9334007d5a67
How to cite
Compositional Multilingual and Behavioral Attribute Steering — arXiv:2609.08410 — arXiv (en), https://arxiv.org/abs/2609.08410 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_9334007d5a67 (Metraipolis archive dc_9334007d5a67, sha256 1abfe5e9902febd7c4b5c680c99db146e67ce4c0fac7a60599f81de535c28bb5)
BibTeX
@misc{dc9334007d5a67,
title = {Compositional Multilingual and Behavioral Attribute Steering — arXiv:2609.08410},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.08410},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_9334007d5a67; sha256 1abfe5e9902febd7c4b5c680c99db146e67ce4c0fac7a60599f81de535c28bb5},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.08410. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
This study examines the compositionality of steering vectors for language and behavioral control in large language models. Focusing on language, jailbreak, and conciseness, we investigate whether additive, training-free composition of attribute steering vectors can preserve the intended steering effect of each attribute, across four instruction-tuned models from two model families and two size scales. We find that single-attribute steering is reliable for all three attributes, but only within an appropriate combination of intervention layer and steering strength, with abstract behaviors (jailbreak, conciseness) favoring middle layers and language favoring earlier layers. We show that additive composition of two attribute vectors succeeds in steering both attributes simultaneously when each is injected at its own best-performing layer, and that this partially extends to three simultaneously composed attributes, addressing an inconsistency left open by prior work on training-free composition. We further analyze the geometric properties of these steering vectors, finding that they are approximately orthogonal in the residual stream, consistent with their compositional behavior. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-09 in cs.CL). Categorie arXiv: Computation and Language (cs.CL). Pagina del lavoro: https://arxiv.org/abs/2609.08410 — 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 2026-10-16T12:30:25.179Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-09?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).