Metraipolis › archive › dc_410ea7bf2b58
VERS: Steering Reasoning Models by Editing Their Activation Explanations — arXiv:2611.03040
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.03040
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-03T18:23:48.654Z
- SHA-256 of the archived text
c58c8a9f3c83e2586976a9f5a9b67665602c9efc27606a79f6a3026526de1ba0- Stable URL of this page
https://metraipolis.com/documento/dc_410ea7bf2b58
How to cite
VERS: Steering Reasoning Models by Editing Their Activation Explanations — arXiv:2611.03040 — arXiv (en), https://arxiv.org/abs/2611.03040 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_410ea7bf2b58 (Metraipolis archive dc_410ea7bf2b58, sha256 c58c8a9f3c83e2586976a9f5a9b67665602c9efc27606a79f6a3026526de1ba0)
BibTeX
@misc{dc410ea7bf2b58,
title = {VERS: Steering Reasoning Models by Editing Their Activation Explanations — arXiv:2611.03040},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.03040},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_410ea7bf2b58; sha256 c58c8a9f3c83e2586976a9f5a9b67665602c9efc27606a79f6a3026526de1ba0},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.03040. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Reasoning language models write a chain of thought before they answer, and many checks in it do not change the final answer. Natural language autoencoders (NLAs) explain a model's activations in text and rebuild activations from text, and editing an explanation can steer a model at a single token. We study whether such edits can also steer how a model reasons. We build VERS (Verbalize, Edit, Reconstruct, Steer): it turns explanation edits at the doubt points of one reasoning trace into one fixed direction, and adds this direction at paragraph breaks. On the full GSM8K test, VERS cuts generated tokens by 45.8% at an accuracy change of -0.82 percentage points (pp); on MATH-500 it saves 31.3% at -3.05 pp. At half strength, its trade-off between tokens and accuracy lies between those of the published methods SEAL and DEER. With fitted affine maps and a strength tuned for each model, the direction transfers to larger Qwen3 models, up to 235B-A22B, and saves 30-35% of tokens. On competition problems, however, VERS loses 3-7 pp at every model size, and a fixed direction did not separate necessary checks from unneeded ones. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno). Categorie arXiv: Computation and Language (cs.CL) ; Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2611.03040 — 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 2026-11-03T18:10:07.987Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).