← Back to all news
arXiv AI By Taha Entesari, Jingyu Zhang, Daniel Khashabi, Mahyar Fazlyab

Minimally Invasive Steering of Language Models

Read the original on arXiv AI →

The paper introduces Minimally Invasive Steering Vector Optimization (MISVO), a method that adjusts a frozen language model’s final hidden states by adding vectors to steer outputs toward a test‑time reward while penalizing changes using the local KL geometry of the token distribution. MISVO derives an analytic Fisher term and a suffix score‑function term, showing that the suffix term is second‑order and that Fisher surrogates match the full KL gradient to first order. Experiments on preference and code‑generation tasks with 1B–14B parameter models demonstrate that MISVO achieves the highest mean reward in most settings while maintaining diversity and coherence comparable to Best‑of‑N.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.

Related stories

arXiv Machine Learning
Jun 15

Towards Steering without Sacrifice: Principled Training of Steering Vectors for Prompt-only Interventions

arXiv:2605. 05983v2 Announce Type: replace Abstract: Recently, steering vectors (SVs) have emerged as an effective and lightweight approach to steer behaviors of large language models (LLMs), among which fine-tuned SVs are more effective than optimization-free ones.

By Yuntai Bao, Qinfeng Li, Xinyan Yu, Ge Su, Wenqi Zhang, Liu Yan, Haiqin Weng, Jianwei Yin, Xuhong Zhang
llmsfine-tuningsafety
More like this →
arXiv Machine Learning
Jun 11

When is Your LLM Steerable?

arXiv:2606. 11599v1 Announce Type: cross Abstract: Activation steering offers a lightweight approach to control language models' behavior at inference time, but whether it succeeds or fails heavily depends on the prompt, concept, model, and steering configuration.

By Chenrui Fan, Yize Cheng, Ming Li, Soheil Feizi, Tianyi Zhou
llms
More like this →
arXiv Machine Learning
Jul 8

A Coin Flip Per Token: Bernoulli Sparse Steering of Large Language Models

arXiv:2607. 05615v1 Announce Type: new Abstract: Activation steering via sparse autoencoders (SAEs) enables behavioral control of large language models without task-specific fine-tuning, but standard methods apply the steering signal at every generated token, incurring constant per-token perturbation that risks degrading fluency.

By Nima Eshraghi, Lovedeep Gondara, Yuqing Huang, Sagarika Suresh, Leizer Teran, Jithin Pradeep, Xiaotong Xu, Fanny Chevalier
llmsreinforcement-learningfine-tuning
More like this →
Hugging Face Trending Papers
Jun 10

When is Your LLM Steerable?

Activation steering offers a lightweight approach to control language models' behavior at inference time, but whether it succeeds or fails heavily depends on the prompt, concept, model, and steering configuration. Finding the regime and boundaries of successful steering typically requires expensive grid searches and post-hoc evaluation of full autoregressive rollouts.

llms
More like this →
arXiv Computation and Language
3d ago

MetaSteer: Context-Conditioned, nonlinear Steering via Attention-Projection Adaptation

MetaSteer is a new method for steering large language models that learns nonlinear, context-dependent interventions applied to attention projection matrices. Unlike traditional linear, context-independent techniques, MetaSteer adapts its effects based on the input, requiring no linear concept-geometry assumption. Trained once on a pooled preference corpus, it transfers zero‑shot to unseen concepts and out‑of‑distribution contexts, matching or surpassing strong task‑specific baselines on multiple benchmarks and model families.

By Mehdi Jafari, Hao Xue, Flora Salim
llmsagentsbenchmarkssafety
More like this →