Adversarial Concept Search: Predicting Compositional Errors From Feature Geometry
cs.AI
Submitted: 2026-06-11
Updated: 2026-11-15
Terminology
Sources
- Toy Models of Superposition
- Decoding by Linear Programming
- Sparse Autoencoders Find Highly Interpretable Features in Language Models
- The Llama 3 Herd of Models
- How Do Language Models Compose Functions?
- Language-Specific Latent Process Hinders Cross-Lingual Performance
- Align Once, Benefit Multilingually: Enforcing Multilingual Consistency for LLM Safety Alignment
- Paths Not Taken: Understanding and Mending the Multilingual Factual Recall Pipeline
- Beyond the Rosetta Stone: Unification Forces in Generalization Dynamics
- The Geometry of Multilingual Language Model Representations
- Does Data Scaling Lead to Visual Compositional Generalization?
- Intriguing properties of neural networks
- Generative Adversarial Networks
- Adversarial Examples Are Not Bugs, They Are Superposition
- Subliminal Effects in Your Data: A General Mechanism via Log-Linearity
- Mechanistic?
- Can Interpretation Predict Behavior on Unseen Data?
- When Models Manipulate Manifolds: The Geometry of a Counting Task
- Internal Causal Mechanisms Robustly Predict Language Model Out-of-Distribution Behaviors
- Diagnosing Generalization Failures from Representational Geometry Markers
Related papers
- Multi-Objective Structured Pruning of LLMs for Latency and Model Size Optimization
- 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
- Embeddings for Preferences, Not Semantics
- Can We Trust Item Response Theory for AI Evaluation?
- Diagnosing Task Insensitivity in Language Agents