Cover Image

Understanding Unreliability of Steering Vectors in Language Models: Geometric Predictors and the Limits of Linear Approximations

Fuente: arXiv

Saved in:

Bibliographic Details
Main Author:	Braun, Joschka
Format:	Preprint
Published:	2026
Subjects:	Computation and Language Artificial Intelligence Machine Learning
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Beyond Multiple Choice: Evaluating Steering Vectors for Summarization
by: Braun, Joschka, et al.
Published: (2025)

Do Personality Traits Interfere? Geometric Limitations of Steering in Large Language Models
by: Bhandari, Pranav, et al.
Published: (2026)

Beyond Linear Steering: Unified Multi-Attribute Control for Language Models
by: Oozeer, Narmeen, et al.
Published: (2025)

That's Deprecated! Understanding, Detecting, and Steering Knowledge Conflicts in Language Models for Code Generation
by: Bae, Jaesung, et al.
Published: (2025)

Compositional Steering of Large Language Models with Steering Tokens
by: Radevski, Gorjan, et al.
Published: (2026)

All or None: Identifiable Linear Properties of Next-token Predictors in Language Modeling
by: Marconato, Emanuele, et al.
Published: (2024)

Steer2Adapt: Dynamically Composing Steering Vectors Elicits Efficient Adaptation of LLMs
by: Han, Pengrui, et al.
Published: (2026)

Improving Steering Vectors by Targeting Sparse Autoencoder Features
by: Chalnev, Sviatoslav, et al.
Published: (2024)

Word Embeddings Are Steers for Language Models
by: Han, Chi, et al.
Published: (2023)

Understanding and Steering the Cognitive Behaviors of Reasoning Models at Test-Time
by: Zhang, Zhenyu, et al.
Published: (2025)

Shifting Perspectives: Steering Vectors for Robust Bias Mitigation in LLMs
by: Siddique, Zara, et al.
Published: (2025)

Linear Representation Transferability Hypothesis: Leveraging Small Models to Steer Large Models
by: Bello, Femi, et al.
Published: (2025)

Endogenous Resistance to Activation Steering in Language Models
by: McKenzie, Alex, et al.
Published: (2026)

Understanding and Mitigating Dataset Corruption in LLM Steering
by: Anderson, Cullen, et al.
Published: (2026)

COLD-Steer: Steering Large Language Models via In-Context One-step Learning Dynamics
by: Sharma, Kartik, et al.
Published: (2026)

Steering Large Language Models for Machine Translation Personalization
by: Scalena, Daniel, et al.
Published: (2025)

FineSteer: A Unified Framework for Fine-Grained Inference-Time Steering in Large Language Models
by: Weng, Zixuan, et al.
Published: (2026)

Towards Understanding Multi-Round Large Language Model Reasoning: Approximability, Learnability and Generalizability
by: Xu, Chenhui, et al.
Published: (2025)

Improving Instruction-Following in Language Models through Activation Steering
by: Stolfo, Alessandro, et al.
Published: (2024)

Multi-Attribute Steering of Language Models via Targeted Intervention
by: Nguyen, Duy, et al.
Published: (2025)

Reasoning or Memorization? Unreliable Results of Reinforcement Learning Due to Data Contamination
by: Wu, Mingqi, et al.
Published: (2025)

In-context Vectors: Making In Context Learning More Effective and Controllable Through Latent Space Steering
by: Liu, Sheng, et al.
Published: (2023)

ILRR: Inference-Time Steering Method for Masked Diffusion Language Models
by: Avrahami, Eden, et al.
Published: (2026)

ContextFocus: Activation Steering for Contextual Faithfulness in Large Language Models
by: Anand, Nikhil, et al.
Published: (2026)

Multi-property Steering of Large Language Models with Dynamic Activation Composition
by: Scalena, Daniel, et al.
Published: (2024)

Interpretable Steering of Large Language Models with Feature Guided Activation Additions
by: Soo, Samuel, et al.
Published: (2025)

Tool Preferences in Agentic LLMs are Unreliable
by: Faghih, Kazem, et al.
Published: (2025)

Steering Without Breaking: Mechanistically Informed Interventions for Discrete Diffusion Language Models
by: Zhou, Hanhan, et al.
Published: (2026)

LLMs Don't Know Their Own Decision Boundaries: The Unreliability of Self-Generated Counterfactual Explanations
by: Mayne, Harry, et al.
Published: (2025)

Picky LLMs and Unreliable RMs: An Empirical Study on Safety Alignment after Instruction Tuning
by: Li, Guanlin, et al.
Published: (2025)

ShadowLLM: Predictor-based Contextual Sparsity for Large Language Models
by: Akhauri, Yash, et al.
Published: (2024)

SAIF: A Sparse Autoencoder Framework for Interpreting and Steering Instruction Following of Language Models
by: He, Zirui, et al.
Published: (2025)

HyperSteer: Activation Steering at Scale with Hypernetworks
by: Sun, Jiuding, et al.
Published: (2025)

Neurocache: Efficient Vector Retrieval for Long-range Language Modeling
by: Safaya, Ali, et al.
Published: (2024)

On the Geometric Structure of Layer Updates in Deep Language Models
by: Yoo, Jun-Sik
Published: (2026)

Understanding (Un)Reliability of Steering Vectors in Language Models
by: Braun, Joschka, et al.
Published: (2025)

Steer Like the LLM: Activation Steering that Mimics Prompting
by: Heyman, Geert, et al.
Published: (2026)

Steering into New Embedding Spaces: Analyzing Cross-Lingual Alignment Induced by Model Interventions in Multilingual Language Models
by: Sundar, Anirudh, et al.
Published: (2025)

Large Language Models as Computable Approximations to Solomonoff Induction
by: Wan, Jun, et al.
Published: (2025)

Theoretical Benefit and Limitation of Diffusion Language Model
by: Feng, Guhao, et al.
Published: (2025)