Metraipolis › archive › dc_7319dd3bcd55
Sampling Reveals Style: Unsupervised, Training-Free Discovery of Prompt-Conditional Stylistic Axes in LLM Activations — arXiv:2609.19150
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.19150
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-10T03:45:50.510Z
- SHA-256 of the archived text
e075ffbc7f0361bbd685621befb44f87172cdd23f8eb4a22a2a6a7802422b372- Stable URL of this page
https://metraipolis.com/documento/dc_7319dd3bcd55
How to cite
Sampling Reveals Style: Unsupervised, Training-Free Discovery of Prompt-Conditional Stylistic Axes in LLM Activations — arXiv:2609.19150 — arXiv (en), https://arxiv.org/abs/2609.19150 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_7319dd3bcd55 (Metraipolis archive dc_7319dd3bcd55, sha256 e075ffbc7f0361bbd685621befb44f87172cdd23f8eb4a22a2a6a7802422b372)
BibTeX
@misc{dc7319dd3bcd55,
title = {Sampling Reveals Style: Unsupervised, Training-Free Discovery of Prompt-Conditional Stylistic Axes in LLM Activations — arXiv:2609.19150},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.19150},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_7319dd3bcd55; sha256 e075ffbc7f0361bbd685621befb44f87172cdd23f8eb4a22a2a6a7802422b372},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.19150. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Large language models (LLMs) encode rich stylistic structure in their hidden activations, but discovering which stylistic dimensions are salient for a given prompt typically requires supervised contrastive data. We present a training-free, prompt-conditional alternative: we repeatedly sample completions of a single prompt at elevated temperature, apply Principal Component Analysis (PCA) to the pooled hidden activations, and label the resulting axes automatically from the pole generations. We validate the discovered axes against 245 human-elicited stylistic annotations in a two-phase study. On our strongest model (Qwen-3.5-4B-Instruct), the top two axes match spontaneously requested human dimensions with 72.8% precision and 43.6% macro-recall, and 75.6% of validity ratings judge the axes' polar generations accurate to their labels, with 90.9% adjacent inter-annotator agreement. Discoverability is strongly model-dependent: both Qwen models and Llama-3.2-3B expose human-salient axes, while DeepSeek-7B-Chat drops to 35.3% precision, its leading components dominated by structural rather than stylistic variance. Simple PCA over a model's own decoding variance is thus an effective, low-cost probe of stylistic structure in LLM representations, one that also exposes sharp cross-model differences in how that structure is organized. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-18 in cs.CL). Categorie arXiv: Computation and Language (cs.CL) ; Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2609.19150 — 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 2026-10-10T03:24:08.103Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-18?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).