Metraipolis › archive › dc_9bdbe5bcf8ff
STR: Supervised Transcoder Replacement for Reducing Steering Side Effects — arXiv:2609.32519
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.32519
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T20:55:38.180Z
- SHA-256 of the archived text
b01fde59a92d5f94b9a50dadf8cf59ca016a6e68fbf810f72c2d2b62bcb719a0- Stable URL of this page
https://metraipolis.com/documento/dc_9bdbe5bcf8ff
How to cite
STR: Supervised Transcoder Replacement for Reducing Steering Side Effects — arXiv:2609.32519 — arXiv (en), https://arxiv.org/abs/2609.32519 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_9bdbe5bcf8ff (Metraipolis archive dc_9bdbe5bcf8ff, sha256 b01fde59a92d5f94b9a50dadf8cf59ca016a6e68fbf810f72c2d2b62bcb719a0)
BibTeX
@misc{dc9bdbe5bcf8ff,
title = {STR: Supervised Transcoder Replacement for Reducing Steering Side Effects — arXiv:2609.32519},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.32519},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_9bdbe5bcf8ff; sha256 b01fde59a92d5f94b9a50dadf8cf59ca016a6e68fbf810f72c2d2b62bcb719a0},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.32519. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Model steering can strengthen a target behavior while degrading other useful behaviors. We introduce Supervised Transcoder Replacement (STR) to reduce these side effects for existing steering methods, including those fitted without a protection objective. STR learns a replacement for the multilayer perceptron (MLP) computation at the steering layer through supervision for target control, non-target preservation, and fidelity without steering. Selected steering methods then fit directions on the frozen replacement while retaining their own fitting objectives. We evaluate three steering methods across Gemma and Llama models using Corrigibility preferences and four harmful-request safety datasets. SALAD-Bench supplies protection training data and a separate in-distribution evaluation split; HarmBench, AdvBench, and StrongREJECT are reserved for out-of-distribution testing. STR substantially reduces steering side effects on the in-distribution evaluation and extends this protection to the unseen safety datasets while retaining effective target control. For target-only supervised steering vectors, pooled out-of-distribution attack success rate falls from 42.46% to 14.42% on Gemma-3-4B and from 34.97% to 12.91% on Gemma-3-12B. These results show that replacement training can benefit steering methods fitted without protection objectives. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI). Categorie arXiv: Artificial Intelligence (cs.AI) ; Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2609.32519 — 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 2026-10-07T20:29:00.845Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).