Metraipolis › archive › dc_e2ad286eafe5
The First Few Tokens Matter: Understanding Chain-of-Thought Controllability — arXiv:2610.40939
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.40939
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-03T05:19:26.144Z
- SHA-256 of the archived text
e9d792b8ef2c245ce469cf6132a2e5b5aab83667999dc5499beb644754bf490d- Stable URL of this page
https://metraipolis.com/documento/dc_e2ad286eafe5
How to cite
The First Few Tokens Matter: Understanding Chain-of-Thought Controllability — arXiv:2610.40939 — arXiv (en), https://arxiv.org/abs/2610.40939 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_e2ad286eafe5 (Metraipolis archive dc_e2ad286eafe5, sha256 e9d792b8ef2c245ce469cf6132a2e5b5aab83667999dc5499beb644754bf490d)
BibTeX
@misc{dce2ad286eafe5,
title = {The First Few Tokens Matter: Understanding Chain-of-Thought Controllability — arXiv:2610.40939},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.40939},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_e2ad286eafe5; sha256 e9d792b8ef2c245ce469cf6132a2e5b5aab83667999dc5499beb644754bf490d},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.40939. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Chain-of-thought (CoT) monitoring is a promising way to oversee increasingly capable agents. It succeeds in part because current models appear limited in their ability to control what they verbalize in their reasoning traces. We investigate the source of this limitation and link it to an entropy collapse in the first few tokens following the opening thinking tag. We exploit this mechanism to recover CoT controllability through two inference-time interventions on initial tokens: prefilling, where even a single token can substantially increase controllability, and banning frequent opening tokens combined with in-context examples, which improves average compliance by 34-41 percentage points across models. These interventions recover mainly addition and style controllability (adding or transforming information), while suppression controllability (avoiding or removing information) exhibits more varied responses, suggesting distinct underlying mechanisms. Consistent with this, the two groups behave as separate capabilities: across models, performance correlates positively within these groups, while cross-group correlations are mostly negative. Importantly, we find that the relationship with monitorability depends on the evaluation: addition and style controllability more consistently positively correlate with monitorability, while suppression often shows negative associations. Our findings motivate the development of controllability evaluations that distinguish difficulty initiating a specific opening from general limitations of the ability to alter reasoning content, alongside separate reporting of control tasks and categories. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2610.40939 — 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 2026-11-03T05:18:52.074Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).