Metraipolis › archive › dc_4ed4e45be28c

Sleight of Word Benchmark: Can Language Models Notice If Their Own Output Was Tampered With? — arXiv:2608.29921

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.29921
License (as archived)
CC0-1.0
Archived
2026-10-17T07:52:18.353Z
SHA-256 of the archived text
f053362f7d975c818010f610bf70690be0ce1da60a80486dce278a685ec4827f
Stable URL of this page
https://metraipolis.com/documento/dc_4ed4e45be28c

How to cite

Sleight of Word Benchmark: Can Language Models Notice If Their Own Output Was Tampered With? — arXiv:2608.29921 — arXiv (en), https://arxiv.org/abs/2608.29921 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_4ed4e45be28c (Metraipolis archive dc_4ed4e45be28c, sha256 f053362f7d975c818010f610bf70690be0ce1da60a80486dce278a685ec4827f)
BibTeX
@misc{dc4ed4e45be28c,
  title = {Sleight of Word Benchmark: Can Language Models Notice If Their Own Output Was Tampered With? — arXiv:2608.29921},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.29921},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_4ed4e45be28c; sha256 f053362f7d975c818010f610bf70690be0ce1da60a80486dce278a685ec4827f},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.29921. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

The output of a Language Model can be tampered with \emph{while} the model is writing it. A simple test can thus be constructed by evaluating the model's perception of this external perturbation. In this spirit, a simple benchmark is built in which a single word is consistently substituted with another in the generation process. We call this method \emph{Sleight of Word}. Two distinct axes are measured: metrics that relate to the model's surprise, as well as an evaluation of the textual reaction for 19 different open-weight language models.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-01 in cs.AI).
Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2608.29921 — 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 2026-10-17T07:28:26.804Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).