Metraipolis › archive › dc_feff68293100

When the Wrong Key Wins: Understanding and Detecting Hallucinations in LLMs — arXiv:2609.15106

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.15106
License (as archived): CC0-1.0
Archived: 2026-10-15T23:15:51.945Z
SHA-256 of the archived text: b27cf039007319ad78e2579648c1d81b0339e5f63349ffa7e8e7b8a84de6e842
Stable URL of this page: https://metraipolis.com/documento/dc_feff68293100

How to cite

When the Wrong Key Wins: Understanding and Detecting Hallucinations in LLMs — arXiv:2609.15106 — arXiv (en), https://arxiv.org/abs/2609.15106 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_feff68293100 (Metraipolis archive dc_feff68293100, sha256 b27cf039007319ad78e2579648c1d81b0339e5f63349ffa7e8e7b8a84de6e842)

BibTeX

@misc{dcfeff68293100,
  title = {When the Wrong Key Wins: Understanding and Detecting Hallucinations in LLMs — arXiv:2609.15106},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.15106},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_feff68293100; sha256 b27cf039007319ad78e2579648c1d81b0339e5f63349ffa7e8e7b8a84de6e842},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.15106. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Large language models can hallucinate even when the knowledge required for a correct answer is already available. We study this failure through a latent-key view of inference, where answer selection depends on competition among associations acquired during pretraining. We show that model predictions can be highly sensitive to individual query keywords, that these influential keywords exhibit entity-specific binding, and that their effects are systematically shaped by pretraining frequency. Multiple bindings can also compete and exhibit higher-order interactions within the same query. Based on this mechanism, we introduce a two-stage keyword-perturbation method for hallucination detection. By removing influential keywords and measuring how the model reorganizes its prediction, the method distinguishes errors caused by misleading key associations from correct decisions supported by diagnostic evidence. Across multiple models and benchmarks, perturbation provides a strong and transferable detection signal, reaching $0.910$ AUROC on probe-known ScientistQA. Finally, we extend the same probabilistic framework to four hallucination regimes: knowledge deficit, wrong knowledge, context distraction, and unstable inference. Their operational distributions across benchmarks provide diagnostic context for why different detector families succeed in different settings.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-15 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2609.15106 — 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 2026-10-15T22:51:09.931Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-15?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).