Metraipolis › archive › dc_aeee7455af90
Hidden Commitment: When Language Models Silently Pick a Side and How Steering Can Surface It — arXiv:2610.38296
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.38296
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-01T02:40:23.614Z
- SHA-256 of the archived text
0ee7552896dc2dbabab68b3b9f2eeb81f7123cc587c791e8621903eed29e6d20- Stable URL of this page
https://metraipolis.com/documento/dc_aeee7455af90
How to cite
Hidden Commitment: When Language Models Silently Pick a Side and How Steering Can Surface It — arXiv:2610.38296 — arXiv (en), https://arxiv.org/abs/2610.38296 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_aeee7455af90 (Metraipolis archive dc_aeee7455af90, sha256 0ee7552896dc2dbabab68b3b9f2eeb81f7123cc587c791e8621903eed29e6d20)
BibTeX
@misc{dcaeee7455af90,
title = {Hidden Commitment: When Language Models Silently Pick a Side and How Steering Can Surface It — arXiv:2610.38296},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.38296},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_aeee7455af90; sha256 0ee7552896dc2dbabab68b3b9f2eeb81f7123cc587c791e8621903eed29e6d20},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.38296. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Language-model answers are often treated as a visible record of the internal decision that produced them. We test this assumption in conflicting-context question answering, where two supplied documents support incompatible answers. A model can process the disagreement and still return a one-sided answer, even under prompts that ask it to disclose conflict. In Qwen 0.8B, hidden source commitment is captured by a residual source-choice direction whose sign tracks which document the answer uses. Sparse middle-layer MLP writers feed the same coordinate, and steering the coordinate changes the chosen source. A transfer run on Gemma 3 1B preserves the source-choice readout and sparse-write bridge under LLM-audited labels. Conflict disclosure follows a different pattern, appearing weaker, later-layer, and only partially steerable. These results support a practical monitoring strategy for retrieval systems that detects silent source commitment with residual readouts and pairs conflict-surfacing prompts with disclosure steering when the model has selected a side under disagreement. Tema degli esploratori: statistica e scienza dei dati (elenco arXiv cs.CV, nuovi lavori del giorno). Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV). Pagina del lavoro: https://arxiv.org/abs/2610.38296 — 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 2026-11-01T02:16:43.877Z dalla pagina https://arxiv.org/list/cs.CV/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).