Steering Interference Reflects the Model's Defaults, Not the Behavior Directions

• URL: http://arxiv.org/abs/2609.06951v1
• Date: Mon, 07 Sep 2026 02:45:13 GMT
• Title: Steering Interference Reflects the Model's Defaults, Not the Behavior Directions
• Abstract summary: Activation steering promises modular control of language model behavior.<n>We ask what decides which other behaviors move, and by how much.<n>We find that it is the model rather than the behavior being steered.
• Score: 14.170453340251981
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Activation steering promises modular control of language model behavior: a behavior such as politeness corresponds to a direction in a model's activations, and adding that direction while it generates should switch the behavior on and leave everything else alone. It does not. We ask what decides which other behaviors move, and by how much, and find that it is the model rather than the behavior being steered. A steer relaxes the model toward a small set of behaviors it already favors, chiefly refusal, sycophancy, and poeticism, and that set is much the same whatever is steered. Three results across 24 behaviors and ten instruction-tuned models support this, every effect read off the generated text by a language-model judge rather than off a probe. That readout matters: all 24 behaviors are linearly decodable, but only 20 change what the model writes. First, a direction carrying no behavioral content, matched to a real steer only in the size of the vector it adds, moves the same behaviors in the same order as real steers do, while producing none of the behaviors that need a specific direction. Second, most interference runs one way, so it cannot be an overlap between two directions: steering profanity makes the model toxic, while steering toxicity leaves profanity untouched. Third, with a behavior held out entirely, geometry measured on the others explains almost none of the interference it takes part in. The account holds on all ten models, the pull toward defaults strongest below 10B parameters and weakening in each family's largest. Reading a steer as a perturbation whose endpoint the model fixes implies that disentangling behavior directions cannot by itself make steering modular.

Related papers

• Amplified Does Not Mean Predictive: Reasoning Behaviors in Thinking Models [56.88565331713778]
  reasoning-oriented training can make traces look more deliberative without amplifying the behaviors most tied to model correctness.<n>We quantify this mismatch with Behavioral Lift, a metric that measures how much correctness changes when a behavior is present versus absent in a model's reasoning trace.<n>We find evidence for an Amplification-Lift Gap, in which thinking models strongly amplify self-awareness, hypothesis testing, and uncertainty acknowledgment.
  arXiv  Detail & Related papers  (2026-08-13T20:37:59Z)
• Divergent Response Modes in Frontier Language Models Under Steering Pressure [0.0]
  This study evaluates behavioral steerability across six frontier models from six developers using 300 paired base and steered items over three categories: values-conflict, reasoning-elicitation, and reasoning-suppression.<n>We find that models differ not just in how much steering shifts their behavior but in what kind of response they give, and some response modes appear in only one or two of them.
  arXiv  Detail & Related papers  (2026-08-06T20:48:14Z)
• They Infer What You Meant: Models Represent Communicative Intent More Reliably Than They Act On It [0.0]
  We show that the failure of a language model to decode the sender's intent is one of readouts on top of a robust representation.<n>A linear probe decodes the sender's intent, whether they want a thing recognized or evaluated, across six models and four families and in the base checkpoints.<n>The behavioral half of the story is established on the recognize/evaluate contrast, where what varies is whether the default output acts on the intent.
  arXiv  Detail & Related papers  (2026-07-03T20:49:37Z)
• Perfect Detection, Failed Control: The Geometry of Knowing vs. Steering in Language Models [1.9247157750972368]
  We test what is the angle between the direction that best detects a behavior and the one that best causes it?<n>If detection implies control the cosine is near 1; otherwise it quantifies a detection gap.<n>The cosine is a weight-computable signature of the dissociation between knowing and steering, not a predictor of it.
  arXiv  Detail & Related papers  (2026-06-23T08:07:51Z)
• Pando: Do Interpretability Methods Work When Models Won't Explain Themselves? [53.07826484214082]
  We introduce Pando, a model-organism benchmark that breaks this confound via an explanation axis.<n>Pando predicts held-out model decisions from $10$ labeled query-response pairs.<n>When explanations are faithful, black-box elicitation matches or exceeds all white-box methods.
  arXiv  Detail & Related papers  (2026-04-13T06:42:24Z)
• What Drives Representation Steering? A Mechanistic Case Study on Steering Refusal [53.189667624047416]
  We show that different steering methodologies leverage functionally interchangeable circuits when applied at the same layer.<n>We show that steering vectors can be sparsified by up to 90-99% while retaining most performance.
  arXiv  Detail & Related papers  (2026-04-09T17:57:14Z)
• Reliable Control-Point Selection for Steering Reasoning in Large Language Models [28.288321095634128]
  Steering vectors offer a training-free mechanism for controlling reasoning behaviors in large language models.<n>However, constructing effective vectors requires identifying genuine behavioral signals in the model's hidden states.<n>We show that current methods detect these behaviors through keyword matching in chain-of-thought traces, implicitly assuming that every detected boundary encodes a genuine behavioral signal.<n>We develop a probabilistic model that formalizes intrinsic reasoning behaviors as events with context-dependent trigger probabilities, and show that unstable boundaries dilute the steering signal.
  arXiv  Detail & Related papers  (2026-04-02T14:48:56Z)
• Understanding Unreliability of Steering Vectors in Language Models: Geometric Predictors and the Limits of Linear Approximations [0.0]
  I investigate why steering reliability differs across behaviors and how it is impacted by steering vector training data.<n>I find that higher cosine similarity between training activation differences predicts more reliable steering.<n>I observe that behavior datasets where positive and negative activations are better separated along the steering direction are more reliably steerable.
  arXiv  Detail & Related papers  (2026-02-19T22:37:05Z)
• GrAInS: Gradient-based Attribution for Inference-Time Steering of LLMs and VLMs [56.93583799109029]
  GrAInS is an inference-time steering approach that operates across both language-only and vision-language models and tasks.<n>During inference, GrAInS hidden activations at transformer layers guided by token-level attribution signals, and normalizes activations to preserve representational scale.<n>It consistently outperforms both fine-tuning and existing steering baselines.
  arXiv  Detail & Related papers  (2025-07-24T02:34:13Z)
• Model Editing with Canonical Examples [75.33218320106585]
  We introduce model editing with canonical examples. A canonical example is a simple instance of good behavior, e.g., The capital of Mauritius is Port Louis. We propose sense finetuning, which selects and finetunes a few sense vectors for each canonical example.
  arXiv  Detail & Related papers  (2024-02-09T03:08:12Z)
• Scaling Laws vs Model Architectures: How does Inductive Bias Influence Scaling? [91.78878523252897]
  This paper conducts a systematic study of scaling behaviour of ten diverse model architectures. We show that architecture is an important consideration when performing scaling and the best performing model can fluctuate at different scales.
  arXiv  Detail & Related papers  (2022-07-21T15:50:22Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.