Active Budget Can Kill Sensitivity: Diagnosing and Repairing TopK Sparse Autoencoder Reliability
cs.AI
Submitted: 2026-09-29
Updated: 2026-09-30
Project page: https://skylion007.github.io/OpenWebTextCorpus
Sources
- Language Model Circuits Are Sparse in the Neuron Basis
- BatchTopK Sparse Autoencoders
- SynthSAEBench: Evaluating Sparse Autoencoders on Scalable Realistic Synthetic Data
- A is for Absorption: Studying Feature Splitting and Absorption in Sparse Autoencoders
- HalluSAE: Detecting Hallucinations in Large Language Models via Sparse Auto-Encoders
- Sparse Autoencoders Find Highly Interpretable Features in Language Models
- Toy Models of Superposition
- Gemma 2: Improving Open Language Models at a Practical Size
- Sanity Checks for Sparse Autoencoders: Do SAEs Beat Random Baselines?
- k-Sparse Autoencoders
- Emergent Linear Representations in World Models of Self-Supervised Sequence Models
- Are LLM Uncertainty and Correctness Encoded by the Same Features? A Functional Dissociation via Sparse Autoencoders
- Qwen2.5 Technical Report
- Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders
- Position: Mechanistic Interpretability Should Prioritize Feature Consistency in SAEs
- Measuring Sparse Autoencoder Feature Sensitivity
- Representation Engineering: A Top-Down Approach to AI Transparency
Related papers
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Agents' Last Exam
- The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
- MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
- Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems