Metraipolis › archive › dc_712edcbc5443

Improving Confidence Estimation in Language Models with Linear Representations of Ambiguity — arXiv:2611.11838

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.11838
License (as archived)
CC0-1.0
Archived
2026-11-10T18:44:43.748Z
SHA-256 of the archived text
1250c0a45a07d365a06ec1733113b437e1cb87c552242e65cf59d89fc668ab7a
Stable URL of this page
https://metraipolis.com/documento/dc_712edcbc5443

How to cite

Improving Confidence Estimation in Language Models with Linear Representations of Ambiguity — arXiv:2611.11838 — arXiv (en), https://arxiv.org/abs/2611.11838 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_712edcbc5443 (Metraipolis archive dc_712edcbc5443, sha256 1250c0a45a07d365a06ec1733113b437e1cb87c552242e65cf59d89fc668ab7a)
BibTeX
@misc{dc712edcbc5443,
  title = {Improving Confidence Estimation in Language Models with Linear Representations of Ambiguity — arXiv:2611.11838},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.11838},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_712edcbc5443; sha256 1250c0a45a07d365a06ec1733113b437e1cb87c552242e65cf59d89fc668ab7a},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.11838. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

As language models (LMs) are increasingly deployed in downstream applications, accurately estimating their likelihood of success on a given example is critical. However, confidence estimates from LMs are poor predictors of their likelihood of success. Accurate confidence estimation is linked in part to the detection of aleatoric uncertainty; we hypothesize that LMs possess representations of at least some aspects of this uncertainty in the form of representations of the inherent ambiguity of the input. Using difference-in-means on minimal pairs of ambiguous and unambiguous sentences, we derive ambiguity representations in LM activations, and find that steering with them selectively (albeit weakly) modulates LMs' judgments of the input's ambiguity. We then propose DART, a fine-tuning method that encourages updates to lie within the span of the ambiguity directions, and penalizes changes to the LM activations outside of this subspace. DART greatly improves calibration, encodes these improvements into the model parameters, and maintains model accuracy on multiple-choice and generative QA tasks.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2611.11838 — 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 2026-11-10T18:37:52.445Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).