Metraipolis › archive › dc_1a3ec43bb367

Legible Failures: Detecting and Repairing In-Context Binding Errors — arXiv:2609.11216

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.11216
License (as archived)
CC0-1.0
Archived
2026-10-16T05:54:55.764Z
SHA-256 of the archived text
f16d1def1d5db432686b90098f5d9b597533ba4461ec19fb56b4ad9a2ade691b
Stable URL of this page
https://metraipolis.com/documento/dc_1a3ec43bb367

How to cite

Legible Failures: Detecting and Repairing In-Context Binding Errors — arXiv:2609.11216 — arXiv (en), https://arxiv.org/abs/2609.11216 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_1a3ec43bb367 (Metraipolis archive dc_1a3ec43bb367, sha256 f16d1def1d5db432686b90098f5d9b597533ba4461ec19fb56b4ad9a2ade691b)
BibTeX
@misc{dc1a3ec43bb367,
  title = {Legible Failures: Detecting and Repairing In-Context Binding Errors — arXiv:2609.11216},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.11216},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_1a3ec43bb367; sha256 f16d1def1d5db432686b90098f5d9b597533ba4461ec19fb56b4ad9a2ade691b},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.11216. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

A wrong answer does not show whether the model lacked the needed information or held it and failed to use it. On an entity-obligation binding task, a language model can emit an incorrect prompt-supplied binding while a linear probe can recover the correct one from its frozen hidden state. We measure how often this occurs across 16 public checkpoints, each evaluated with three seeds. We fit a probe on a training fold, select its layer on a validation fold, and report results on a disjoint test fold. On the trials each model gets wrong, probe accuracy exceeds the strict present-obligation baseline, 1/K = 0.125, by +0.196 (95% CI [+0.101, +0.296], bootstrapped over models). A query-entity counterfactual rules out token presence and recency. A score built from the sign of probe-output disagreement improves failure detection over the model's own confidence by +0.079 AUROC (95% CI [+0.036, +0.126]). Raw probe confidence gives no measurable improvement over model confidence. Steering the residual stream toward the probe-decoded binding, with no gold label, raises accuracy on all eight models tested by a mean of +0.168 (95% CI [+0.066, +0.280]). Where recent studies report that probe-detected errors are resistant to interventions, we find that in-context binding is a setting in which probes are actionable.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-11 in cs.LG).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.11216 — 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 2026-10-16T05:44:24.989Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-09-11?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).