Stop Probing, Start Coding: Why Linear Probes and Sparse Autoencoders Fail at Compositional Generalisation

arXiv:2603.28744 · cs.LG · Submitted 2026-03-30 · Read on arXiv

cs.LG

Submitted: 2026-03-30

Updated: 2026-10-07

Terminology

• Linear probe — 1× in this paper · explained in 2 episodes
• Sparse Autoencoders (SAEs) — 1× in this paper · explained in 3 episodes
• Sparsity — 1× in this paper · explained in 4 episodes

Sources

• Improving Steering Vectors by Targeting Sparse Autoencoder Features
• Sparse Autoencoders Find Highly Interpretable Features in Language Models
• Toy Models of Superposition
• Cooper: A Library for Constrained Optimization in Deep Learning
• Sparse Shift Autoencoders for Identifying Concepts from Large Language Model Activations
• Reducing the Amortization Gap in Variational Autoencoders: A Bayesian Random Function Approach
• From superposition to sparse codes: interpretable representations in neural networks
• Resurrecting the Salmon: Rethinking Mechanistic Interpretability with Domain-Specific Sparse Autoencoders
• Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders
• Position: Adopt Constraints Over Fixed Penalties in Deep Learning
• Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalization
• Steering Language Models With Activation Engineering

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