Guarded Gradient-Based Activation Steering of Shutdown Responses in Qwen3.5-0.8B: A Minimum-Step Policy
cs.LG
Submitted: 2026-09-23
Updated: 2026-09-23
Comments: 9 pages, 2 tables
Code: https://github.com/Farhad-Davaripour/sp_lense
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Understanding (Un)Reliability of Steering Vectors in Language Models
- Plug and Play Language Models: A Simple Approach to Controlled Text Generation
- Verbalizable Representations Form a Global Workspace in Language Models
- Programming Refusal with Conditional Activation Steering
- Quantifying Self-Preservation Bias in Large Language Models
- Peer-Preservation in Frontier Models
- Incomplete Tasks Induce Shutdown Resistance in Some Frontier LLMs
- Steering Language Models With Activation Engineering
- Representation Engineering: A Top-Down Approach to AI Transparency
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks