Are Sparse Autoencoders Useful? A Case Study in Sparse Probing

TL;DR

This study evaluates the usefulness of sparse autoencoders (SAEs) in interpreting large language models by testing their impact on downstream tasks under challenging conditions, revealing limited advantages over simpler baselines.

Contribution

The paper provides a comprehensive empirical assessment of SAEs in real-world probing tasks, highlighting their limitations and the importance of rigorous downstream evaluation.

Findings

SAEs occasionally outperform baselines on individual datasets

Ensemble methods with SAEs do not consistently outperform baseline ensembles

Simple non-SAE baselines achieve similar interpretability results

Abstract

Sparse autoencoders (SAEs) are a popular method for interpreting concepts represented in large language model (LLM) activations. However, there is a lack of evidence regarding the validity of their interpretations due to the lack of a ground truth for the concepts used by an LLM, and a growing number of works have presented problems with current SAEs. One alternative source of evidence would be demonstrating that SAEs improve performance on downstream tasks beyond existing baselines. We test this by applying SAEs to the real-world task of LLM activation probing in four regimes: data scarcity, class imbalance, label noise, and covariate shift. Due to the difficulty of detecting concepts in these challenging settings, we hypothesize that SAEs' basis of interpretable, concept-level latents should provide a useful inductive bias. However, although SAEs occasionally perform better than baselines on individual datasets, we are unable to design ensemble methods combining SAEs with baselines that consistently outperform ensemble methods solely using baselines. Additionally, although SAEs initially appear promising for identifying spurious correlations, detecting poor dataset quality, and training multi-token probes, we are able to achieve similar results with simple non-SAE baselines as well. Though we cannot discount SAEs' utility on other tasks, our findings highlight the shortcomings of current SAEs and the need to rigorously evaluate interpretability methods on downstream tasks with strong baselines.