Train the Model, Not the Reader: Decodability Supervision for Verifiable Activation Explanations
Hiskias Dingeto
cs.AI, cs.CL
Submitted: 2026-08-19
Updated: 2026-08-20
Terminology
Sources
- Faithfulness Tests for Natural Language Explanations
- Obfuscated Activations Bypass LLM Latent-Space Defenses
- Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation
- Building Better Activation Oracles
- Eliciting Latent Predictions from Transformers with the Tuned Lens
- Transferring Linear Features Across Language Models With Model Stitching
- CycleGAN, a Master of Steganography
- Gradient Routing: Masking Gradients to Localize Computation in Neural Networks
- Do Activation Monitors Survive Model Updates? Benchmarking, Predicting, and Repairing Activation-Monitor Staleness
- The Pile: An 800GB Dataset of Diverse Text for Language Modeling
- Introspective Coupling: Self-Explanation Training Tracks Behavioral Change Despite Fixed Supervision
- Rigorously Assessing Natural Language Explanations of Neurons
- Predictive Concept Decoders: Training Scalable End-to-End Interpretability Assistants
- Sparse Autoencoders Find Highly Interpretable Features in Language Models
- Activation Oracles: Training and Evaluating LLMs as General-Purpose Activation Explainers
- Measuring Faithfulness in Chain-of-Thought Reasoning
- Do Activation Verbalization Methods Convey Privileged Information?
- Promises and Pitfalls of Black-Box Concept Learning Models
- Do Concept Bottleneck Models Learn as Intended?
- Hidden in Plain Text: Emergence & Mitigation of Steganographic Collusion in LLMs
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
- 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
- VSAL: A Vision Solver with Adaptive Layouts for Graph Property Detection