Metraipolis › archive › dc_788dd758e33b

Correct Labels Are Not Enough: Reliable Metadata Can Silently Control Model Behavior — arXiv:2611.04395

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.04395
License (as archived)
CC0-1.0
Archived
2026-11-04T12:27:52.586Z
SHA-256 of the archived text
43e71448024af72a121bc3d0efa4835d1843abc6bedd5bf12a56fa8239bc5197
Stable URL of this page
https://metraipolis.com/documento/dc_788dd758e33b

How to cite

Correct Labels Are Not Enough: Reliable Metadata Can Silently Control Model Behavior — arXiv:2611.04395 — arXiv (en), https://arxiv.org/abs/2611.04395 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_788dd758e33b (Metraipolis archive dc_788dd758e33b, sha256 43e71448024af72a121bc3d0efa4835d1843abc6bedd5bf12a56fa8239bc5197)
BibTeX
@misc{dc788dd758e33b,
  title = {Correct Labels Are Not Enough: Reliable Metadata Can Silently Control Model Behavior — arXiv:2611.04395},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.04395},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_788dd758e33b; sha256 43e71448024af72a121bc3d0efa4835d1843abc6bedd5bf12a56fa8239bc5197},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.04395. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Fine-tuning data can include metadata that predicts the correct answer, such as a reward-model score or a quality label. We ask whether a model learns to rely on this metadata even when every training label is correct. We fine-tune 7-9B open-weight models from three families on multiple-choice questions in which (Reward: 5) always marks the correct option; when this annotation is moved to a wrong option at test time, models select that option on 97-100% of CommonsenseQA items. Removing it leaves most answers correct, so evaluation without it misses the dependence. Cue-following rises sharply with training-time reliability and develops even with neutral or nonsense annotation words, although test-time wording affects its strength. A Qwen3 experiment also finds cue dependence in option-letter scores after direct preference optimization. Audits can test for this dependence by moving or removing visible annotations while holding the task and correct answer fixed.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2611.04395 — 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 2026-11-04T12:26:47.152Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).