Metraipolis › archive › dc_ecfb7d0d214f

When Steering Fails in Latent Reasoning: A Latent-to-Language Transition Gap — arXiv:2609.21662

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.21662
License (as archived)
CC0-1.0
Archived
2026-10-10T01:12:46.652Z
SHA-256 of the archived text
b0f39303d284bae1a7b11536e9204b7c40772a0ff12bc4f84b4da9e0b3f079d3
Stable URL of this page
https://metraipolis.com/documento/dc_ecfb7d0d214f

How to cite

When Steering Fails in Latent Reasoning: A Latent-to-Language Transition Gap — arXiv:2609.21662 — arXiv (en), https://arxiv.org/abs/2609.21662 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ecfb7d0d214f (Metraipolis archive dc_ecfb7d0d214f, sha256 b0f39303d284bae1a7b11536e9204b7c40772a0ff12bc4f84b4da9e0b3f079d3)
BibTeX
@misc{dcecfb7d0d214f,
  title = {When Steering Fails in Latent Reasoning: A Latent-to-Language Transition Gap — arXiv:2609.21662},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.21662},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ecfb7d0d214f; sha256 b0f39303d284bae1a7b11536e9204b7c40772a0ff12bc4f84b4da9e0b3f079d3},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.21662. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Activation steering has become a widely used approach for controlling language models during explicit chain-of-thought (CoT) reasoning, motivating its extension to latent CoT. However, we find that steering continuous thoughts produces substantially weaker effects on subsequent language generation than steering explicit CoT, even when the hidden representations are moved by comparable amounts. We first show that task information remains identifiable in continuous thoughts. Hence, we hypothesize a \textbf{latent-to-language transition gap}, in which an intervention effect in latent space fails to transfer to language generation. Two further results support this hypothesis: the output distribution changes abruptly at the transition boundary, and task-related directions exert much weaker bidirectional control in latent CoT than in explicit CoT. These findings identify the transition interface as a central target for evaluating and designing future latent-steering methods.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-21 in cs.AI).
Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.21662 — 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 2026-10-10T00:23:42.555Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-21?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).