Subliminal Learning Is Steering Vector Distillation

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Blank, Camila, Bhatia, Agam, Rajamanoharan, Senthooran, Conmy, Arthur, Nanda, Neel
Format:	Preprint
Published:	2026
Subjects:	Artificial Intelligence
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

How Well Do Models Follow Their Constitutions?
by: Jakkli, Arya, et al.
Published: (2026)

Chain-of-Thought Reasoning In The Wild Is Not Always Faithful
by: Arcuschin, Iván, et al.
Published: (2025)

Improving Dictionary Learning with Gated Sparse Autoencoders
by: Rajamanoharan, Senthooran, et al.
Published: (2024)

Understanding Reasoning in Thinking Language Models via Steering Vectors
by: Venhoff, Constantin, et al.
Published: (2025)

Emergent Misalignment is Easy, Narrow Misalignment is Hard
by: Soligo, Anna, et al.
Published: (2026)

Convergent Linear Representations of Emergent Misalignment
by: Soligo, Anna, et al.
Published: (2025)

Do I Know This Entity? Knowledge Awareness and Hallucinations in Language Models
by: Ferrando, Javier, et al.
Published: (2024)

Steering Out-of-Distribution Generalization with Concept Ablation Fine-Tuning
by: Casademunt, Helena, et al.
Published: (2025)

Thought Branches: Interpreting LLM Reasoning Requires Resampling
by: Macar, Uzay, et al.
Published: (2025)

Model Organisms for Emergent Misalignment
by: Turner, Edward, et al.
Published: (2025)

Are Sparse Autoencoders Useful? A Case Study in Sparse Probing
by: Kantamneni, Subhash, et al.
Published: (2025)

Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2
by: Lieberum, Tom, et al.
Published: (2024)

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

Base Models Know How to Reason, Thinking Models Learn When
by: Venhoff, Constantin, et al.
Published: (2025)

Thought Anchors: Which LLM Reasoning Steps Matter?
by: Bogdan, Paul C., et al.
Published: (2025)

Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders
by: Rajamanoharan, Senthooran, et al.
Published: (2024)

Eliciting Secret Knowledge from Language Models
by: Cywiński, Bartosz, et al.
Published: (2025)

Scaling sparse feature circuit finding for in-context learning
by: Kharlapenko, Dmitrii, et al.
Published: (2025)

Towards eliciting latent knowledge from LLMs with mechanistic interpretability
by: Cywiński, Bartosz, et al.
Published: (2025)

Building Production-Ready Probes For Gemini
by: Kramár, János, et al.
Published: (2026)

Subliminal Transfer of Unsafe Behaviors in AI Agent Distillation
by: Dang, Jacob, et al.
Published: (2026)

Steering Evaluation-Aware Language Models to Act Like They Are Deployed
by: Hua, Tim Tian, et al.
Published: (2025)

Simple Mechanistic Explanations for Out-Of-Context Reasoning
by: Wang, Atticus, et al.
Published: (2025)

Subliminal Learning is a LoRA Artifact
by: Nief, Todd, et al.
Published: (2026)

Dense SAE Latents Are Features, Not Bugs
by: Sun, Xiaoqing, et al.
Published: (2025)

Automatically Finding Reward Model Biases
by: Wang, Atticus, et al.
Published: (2026)

Explorations of Self-Repair in Language Models
by: Rushing, Cody, et al.
Published: (2024)

Towards Best Practices of Activation Patching in Language Models: Metrics and Methods
by: Zhang, Fred, et al.
Published: (2023)

Applying sparse autoencoders to unlearn knowledge in language models
by: Farrell, Eoin, et al.
Published: (2024)

Learning Multi-Level Features with Matryoshka Sparse Autoencoders
by: Bussmann, Bart, et al.
Published: (2025)

When Chain of Thought is Necessary, Language Models Struggle to Evade Monitors
by: Emmons, Scott, et al.
Published: (2025)

BatchTopK Sparse Autoencoders
by: Bussmann, Bart, et al.
Published: (2024)

Towards Understanding Subliminal Learning: When and How Hidden Biases Transfer
by: Schrodi, Simon, et al.
Published: (2025)

Sustained Gradient Alignment Mediates Subliminal Learning in a Multi-Step Setting: Evidence from MNIST Auxiliary Logit Distillation Experiment
by: Kitkana, Chayanon, et al.
Published: (2026)

Hypothetical Minds: Scaffolding Theory of Mind for Multi-Agent Tasks with Large Language Models
by: Cross, Logan, et al.
Published: (2024)

Fluid Representations in Reasoning Models
by: Kharlapenko, Dmitrii, et al.
Published: (2026)

Line of Sight: On Linear Representations in VLLMs
by: Rajaram, Achyuta, et al.
Published: (2025)

Subliminal Learning: Language models transmit behavioral traits via hidden signals in data
by: Cloud, Alex, et al.
Published: (2025)

Learning Through Noise: Why Subliminal Learning Works and When It Fails
by: Brockers, Vincent C., et al.
Published: (2026)

Because we have LLMs, we Can and Should Pursue Agentic Interpretability
by: Kim, Been, et al.
Published: (2025)