AlignSAE: Concept-Aligned Sparse Autoencoders
This addresses the challenge of interpretability and control in LLMs for researchers and practitioners, though it is incremental as it builds on existing sparse autoencoder methods.
The paper tackles the problem of aligning sparse autoencoder features with human-defined concepts in large language models, resulting in AlignSAE, which enables precise causal interventions like concept swaps by targeting semantically aligned slots.
Large Language Models (LLMs) encode factual knowledge within hidden parametric spaces that are difficult to inspect or control. While Sparse Autoencoders (SAEs) can decompose hidden activations into more fine-grained, interpretable features, they often struggle to reliably align these features with human-defined concepts, resulting in entangled and distributed feature representations. To address this, we introduce AlignSAE, a method that aligns SAE features with a defined ontology through a "pre-train, then post-train" curriculum. After an initial unsupervised training phase, we apply supervised post-training to bind specific concepts to dedicated latent slots while preserving the remaining capacity for general reconstruction. This separation creates an interpretable interface where specific relations can be inspected and controlled without interference from unrelated features. Empirical results demonstrate that AlignSAE enables precise causal interventions, such as reliable "concept swaps", by targeting single, semantically aligned slots.