Metraipolis › archive › dc_675f4e3f893f

Stable Miscalibration in Large Language Models: A Practical View of High-Confidence Errors — arXiv:2608.13591

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.13591
License (as archived)
CC0-1.0
Archived
2026-10-19T02:20:15.785Z
SHA-256 of the archived text
b7e789f6b1f56d4f549ba3cf7b6b7de0fa54910cc56e3fc89222f120f5abd916
Stable URL of this page
https://metraipolis.com/documento/dc_675f4e3f893f

How to cite

Stable Miscalibration in Large Language Models: A Practical View of High-Confidence Errors — arXiv:2608.13591 — arXiv (en), https://arxiv.org/abs/2608.13591 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_675f4e3f893f (Metraipolis archive dc_675f4e3f893f, sha256 b7e789f6b1f56d4f549ba3cf7b6b7de0fa54910cc56e3fc89222f120f5abd916)
BibTeX
@misc{dc675f4e3f893f,
  title = {Stable Miscalibration in Large Language Models: A Practical View of High-Confidence Errors — arXiv:2608.13591},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.13591},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_675f4e3f893f; sha256 b7e789f6b1f56d4f549ba3cf7b6b7de0fa54910cc56e3fc89222f120f5abd916},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.13591. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

High-confidence errors in large language models are often treated as evidence of fragile internal inference. We study a different possibility: stable miscalibration, where a confident wrong answer remains locally stable under small perturbations. We combine two diagnostics: a label-aware output-level audit score that ranks domains by confidence variation and overconfident mistakes under a forced-answer baseline, and an internal sensitivity probe that measures hidden-state movement. On a multi-domain binary factual audit set, this audit score tracks where abstention-aware self-critique reduces decision loss, although direct labeled baselines rank the same gain more strongly. Internally, self-critical prompting consistently reduces hidden-state sensitivity across layers in three open-weight models. This supports prompt-induced local stabilization rather than a purely output-level abstention pattern, but it does not imply calibration: audit-defined overconfident errors are not clearly more locally sensitive than confidently correct answers, so some high-confidence errors may be stable and miscalibrated rather than simply fragile.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-17 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2608.13591 — 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 2026-10-19T02:02:02.715Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-08-17?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).