Metraipolis › archive › dc_785bb71baaf9

Activation Flow: Manufacturing Activations for Steering — arXiv:2609.32530

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.32530
License (as archived)
CC0-1.0
Archived
2026-10-07T21:00:20.707Z
SHA-256 of the archived text
893b558cd4c98f40d483fc5a1edaaf117fc27d70bba0eeb76e1dc217794c944d
Stable URL of this page
https://metraipolis.com/documento/dc_785bb71baaf9

How to cite

Activation Flow: Manufacturing Activations for Steering — arXiv:2609.32530 — arXiv (en), https://arxiv.org/abs/2609.32530 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_785bb71baaf9 (Metraipolis archive dc_785bb71baaf9, sha256 893b558cd4c98f40d483fc5a1edaaf117fc27d70bba0eeb76e1dc217794c944d)
BibTeX
@misc{dc785bb71baaf9,
  title = {Activation Flow: Manufacturing Activations for Steering — arXiv:2609.32530},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.32530},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_785bb71baaf9; sha256 893b558cd4c98f40d483fc5a1edaaf117fc27d70bba0eeb76e1dc217794c944d},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.32530. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Difference-in-means steering requires activations recorded while a model shows the desired behavior, which a sandbagging model withholds by deliberately underperforming. We introduce Activation Flow (ActFlow), which manufactures these activations from $k$ correct labels without fine-tuning. ActFlow sets target logits that rank each labeled item's correct answer first, and moves the logits toward them by adding one vector $x$ to all $k$ residual streams at one layer. ActFlow is a family of ordinary differential equations for $x$, one for each rule that maps the required logit change to the velocity of $x$. The smallest-norm rule lands exactly on the targets, while the others keep only the top singular directions of the Jacobian. We test ActFlow on three instruction-tuned models, each locked by a sandbagging prompt and by a password-locked LoRA. At $k=40$, ActFlow keeping five singular directions raises the mean held-out ARC-Easy accuracy over the six locked models from $0.05$ to $0.85$, against $0.88$ for rank-$16$ LoRA fine-tuning, which trains about $3{,}000$ times as many parameters, and $0.92$ for the honest models. Furthermore, it scores higher than the smallest-norm rule in 16 of the 18 combinations of locked model and $k$, and its steering direction is nearly orthogonal to the honest difference-in-means direction. It also unlocks two LoRA locks where the honest direction fails. Surprisingly, with one labeled item, the single-step smallest-norm rule raises the Qwen2.5-7B prompt lock from $0.04$ to $0.59$.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Differential Geometry (math.DG).
Pagina del lavoro: https://arxiv.org/abs/2609.32530 — 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 2026-10-07T20:29:02.369Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).