Metraipolis › archive › dc_fc0dc0a8ca2f

Detecting Hidden Chain-of-Thought in Large Language Models with Linguistic, Behavioral, and Mechanistic Indicators — arXiv:2608.29956

Source: arXiv (en)
Original: https://arxiv.org/abs/2608.29956
License (as archived): CC0-1.0
Archived: 2026-10-17T08:07:17.932Z
SHA-256 of the archived text: 9cc296e8a5a3641c25c73b2f94beb64f0e5402be5838cda255ed83d5252f7717
Stable URL of this page: https://metraipolis.com/documento/dc_fc0dc0a8ca2f

How to cite

Detecting Hidden Chain-of-Thought in Large Language Models with Linguistic, Behavioral, and Mechanistic Indicators — arXiv:2608.29956 — arXiv (en), https://arxiv.org/abs/2608.29956 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_fc0dc0a8ca2f (Metraipolis archive dc_fc0dc0a8ca2f, sha256 9cc296e8a5a3641c25c73b2f94beb64f0e5402be5838cda255ed83d5252f7717)

BibTeX

@misc{dcfc0dc0a8ca2f,
  title = {Detecting Hidden Chain-of-Thought in Large Language Models with Linguistic, Behavioral, and Mechanistic Indicators — arXiv:2608.29956},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.29956},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_fc0dc0a8ca2f; sha256 9cc296e8a5a3641c25c73b2f94beb64f0e5402be5838cda255ed83d5252f7717},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.29956. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Large language models often answer complex reasoning questions without revealing intermediate steps, raising whether they reason latently or complete patterns. We propose the Hidden CoT Detection Score (HCDS), a comparative behavioral and mechanistic signal measuring whether neutral-prompt behavior aligns more closely with explicit CoT or explicit no- CoT. Here, hidden CoT operationally denotes this neutral-prompt CoT-like alignment; HCDS does not directly observe or prove an unexposed reasoning trace. On GSM8K, HCDS is significantly positive for both Qwen3-4B variants (Thinking $+1.87$, $p = 1.2 \times 10^{-7}$; Instruct $+1.41$, $p = 1.9 \times 10^{-4}$), replicates across a different inference stack and quantization within $0.08$ ($+1.80$ and $+1.45$), and is not significantly positive in seven of eight length-adjusted calibration-control cells. The unadjusted score produces large positive scores on single-step arithmetic and numeric factual lookup. The variants also respond differently to no-CoT instructions: Instruct complies from the prompt alone, whereas Thinking continues reasoning and requires intervention. These findings show stronger, less prompt-conditional CoT-like behavior in the reasoning-tuned model, consistent with but not proof of latent reasoning. HCDS thus investigates latent reasoning without relying on models' self-reported traces.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-01 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2608.29956 — 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 2026-10-17T07:47:52.762Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).