Metraipolis › archive › dc_2f212bfbc122
Learning When to Trust via Selective Context Preference Optimization — arXiv:2608.06377
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.06377
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-25T21:27:27.474Z
- SHA-256 of the archived text
4a201087ff88a795160709474975c2569c6434dacbcbe0f766d66ce1b636a848- Stable URL of this page
https://metraipolis.com/documento/dc_2f212bfbc122
How to cite
Learning When to Trust via Selective Context Preference Optimization — arXiv:2608.06377 — arXiv (en), https://arxiv.org/abs/2608.06377 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_2f212bfbc122 (Metraipolis archive dc_2f212bfbc122, sha256 4a201087ff88a795160709474975c2569c6434dacbcbe0f766d66ce1b636a848)
BibTeX
@misc{dc2f212bfbc122,
title = {Learning When to Trust via Selective Context Preference Optimization — arXiv:2608.06377},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.06377},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_2f212bfbc122; sha256 4a201087ff88a795160709474975c2569c6434dacbcbe0f766d66ce1b636a848},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.06377. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Language models increasingly condition their answers on external signals, and a single misleading one can turn a correct answer wrong. The obvious remedy, training models to resist such signals, hides a failure mode: a model that ignores all context looks robust yet is useless when the context is worth trusting. We recast the problem as selective trust and introduce MIST, a human-annotated benchmark that renders each reasoning item under four matched conditions (clean, misleading, correct-context, and irrelevant-context), together with SC2W, a paired metric counting how often a misleading signal flips a clean-correct answer to wrong. Across a comprehensive benchmark study, we observe that such a susceptibility is universal. We then propose SCOPE, which mines clean-correct/misleading-wrong failures and optimizes a standard Direct Preference Optimization (DPO) objective over matched preference pairs balanced equally across all four conditions, rather than over misleading items alone. Our approach substantially reduces SC2W on popular open-sourced models while preserving accuracy when the added context is clean, correct, or irrelevant. With this work, we argue that models should be judged on selective trust, not on resistance alone. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-07 in cs.CL). Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI); Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2608.06377 — 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 2026-10-25T21:24:38.234Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-08-07?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).