Metraipolis › archive › dc_4ccfed7343b8
Off-Target Effects of Response-Style Alignment in a Korean 27B Language Model — arXiv:2609.11291
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.11291
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-16T05:23:53.824Z
- SHA-256 of the archived text
133137a31007c8fbf9a07c95bc8a1fe83febcb72e9824765fb66560107b833a9- Stable URL of this page
https://metraipolis.com/documento/dc_4ccfed7343b8
How to cite
Off-Target Effects of Response-Style Alignment in a Korean 27B Language Model — arXiv:2609.11291 — arXiv (en), https://arxiv.org/abs/2609.11291 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_4ccfed7343b8 (Metraipolis archive dc_4ccfed7343b8, sha256 133137a31007c8fbf9a07c95bc8a1fe83febcb72e9824765fb66560107b833a9)
BibTeX
@misc{dc4ccfed7343b8,
title = {Off-Target Effects of Response-Style Alignment in a Korean 27B Language Model — arXiv:2609.11291},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.11291},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_4ccfed7343b8; sha256 133137a31007c8fbf9a07c95bc8a1fe83febcb72e9824765fb66560107b833a9},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.11291. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
We post-train Qwen3.8-27B for Korean response style -- verbosity, list and markdown usage, discourse structure and register -- and measure two behaviours the objective never targets: abstention on ambiguous social questions in KoBBQ, where the benchmark-correct answer is UNKNOWN, and unprompted disclosure in securities guidance. Both move, and the changes are expressed primarily through the model's emission policy: how often it answers and how much it says. Matched target-form controls show that answer propensity depends on the training target, not the prompt set or recipe alone. Holding prompts, recipe, data volume and serving fixed and changing only the target text, three style seeds give positive answer-rate point estimates (mean +0.82 pp) and three neutral seeds negative ones (mean -1.53 pp); the observed seed ranges do not overlap and the means differ by 2.34 pp. A length-matched arm lies between them, and a fourth arm that stays short while preserving hedging is unstable across seeds, so which feature of the form is responsible is unresolved. For absolute stereotyped exposure the decomposition into an answer-propensity term and a conditional-composition term is an algebraic identity, not a finding; its empirical content is where the movement went. Across the trained checkpoints the changes are dominated by answer propensity while the composition term stays small, and because that term is evaluated on treatment-dependent answered subsets we do not read it as evidence about latent preference. Two measurement results follow. A between-arm contrast in conditional stereotyped share does not identify a change in conditional content preference when answer status is treatment-dependent. And agreement between two rule detectors for the same construct runs from 0.44 to 0.99 depending on which checkpoint produced the text -- observable without any reference labels. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-11 in cs.AI). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.11291 — 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 2026-10-16T05:17:52.001Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-11?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).