Metraipolis › archive › dc_bd219913aca9

Thinking Leakage: A Causal Audit of NoThink Post-Training in Hybrid Reasoning Models — arXiv:2609.28682

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.28682
License (as archived)
CC0-1.0
Archived
2026-10-08T05:25:43.669Z
SHA-256 of the archived text
0c9c8a56844941e1fd03f81127b682beef0e0d58bb3d13f4800d263e35819298
Stable URL of this page
https://metraipolis.com/documento/dc_bd219913aca9

How to cite

Thinking Leakage: A Causal Audit of NoThink Post-Training in Hybrid Reasoning Models — arXiv:2609.28682 — arXiv (en), https://arxiv.org/abs/2609.28682 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_bd219913aca9 (Metraipolis archive dc_bd219913aca9, sha256 0c9c8a56844941e1fd03f81127b682beef0e0d58bb3d13f4800d263e35819298)
BibTeX
@misc{dcbd219913aca9,
  title = {Thinking Leakage: A Causal Audit of NoThink Post-Training in Hybrid Reasoning Models — arXiv:2609.28682},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.28682},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_bd219913aca9; sha256 0c9c8a56844941e1fd03f81127b682beef0e0d58bb3d13f4800d263e35819298},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.28682. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Post-training hybrid reasoning models in NoThink mode has attracted growing interest as a way to improve performance while keeping inference fast. However, these gains may draw on thinking behavior already accessible through the base model's Think mode. We formulate this thinking leakage in a causal mediation framework and audit its contribution using bidirectional interventions along a simple base-derived activation direction. Across three models and three post-training methods on competition math benchmarks, we find that leakage is real, causal, and substantial: behavioral and representational analyses reveal shifts toward Think, steering the base model along this direction reproduces most of the post-training accuracy gain, and counter-steering a checkpoint removes a substantial share of what it gains. Across nine aligned checkpoints with positive NoThink gains, the resulting leakage ratio ranges from 42% to 79%. These interventions support a substantial causal contribution of thinking leakage. Our findings show that a post-training method's apparent advantage can therefore reflect greater drift toward Think, obscuring whether it improves capability within NoThink or more effectively re-invokes existing Think behavior.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-25 in cs.LG).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.28682 — 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 2026-10-08T04:50:39.223Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-09-25?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).