Towards Principled Evaluations of Sparse Autoencoders for Interpretability and Control

TL;DR

This paper introduces a framework for evaluating sparse autoencoders in interpretability tasks by comparing unsupervised features to supervised benchmarks, revealing insights into feature quality and training phenomena.

Contribution

It proposes a novel evaluation framework for sparse autoencoders using supervised dictionaries as ground-truth references, enabling more objective interpretability assessments.

Findings

SAEs capture interpretable features for IOI task

SAEs are less effective than supervised features in controlling models

Identified phenomena: feature occlusion and over-splitting

Abstract

Disentangling model activations into meaningful features is a central problem in interpretability. However, the absence of ground-truth for these features in realistic scenarios makes validating recent approaches, such as sparse dictionary learning, elusive. To address this challenge, we propose a framework for evaluating feature dictionaries in the context of specific tasks, by comparing them against \emph{supervised} feature dictionaries. First, we demonstrate that supervised dictionaries achieve excellent approximation, control, and interpretability of model computations on the task. Second, we use the supervised dictionaries to develop and contextualize evaluations of unsupervised dictionaries along the same three axes. We apply this framework to the indirect object identification (IOI) task using GPT-2 Small, with sparse autoencoders (SAEs) trained on either the IOI or OpenWebText datasets. We find that these SAEs capture interpretable features for the IOI task, but they are less successful than supervised features in controlling the model. Finally, we observe two qualitative phenomena in SAE training: feature occlusion (where a causally relevant concept is robustly overshadowed by even slightly higher-magnitude ones in the learned features), and feature over-splitting (where binary features split into many smaller, less interpretable features). We hope that our framework will provide a useful step towards more objective and grounded evaluations of sparse dictionary learning methods.