Interpretable Embeddings with Sparse Autoencoders: A Data Analysis Toolkit

TL;DR

This paper introduces sparse autoencoder embeddings as an interpretable, cost-effective, and controllable alternative to dense embeddings and LLMs for analyzing large-scale text data, uncovering insights and biases.

Contribution

The paper presents a novel application of sparse autoencoders for creating interpretable embeddings that outperform dense models and LLMs in cost, reliability, and controllability for data analysis tasks.

Findings

SAE embeddings reveal semantic differences between datasets

SAE embeddings uncover unexpected concept correlations

SAE embeddings outperform dense embeddings in property-based retrieval

Abstract

Analyzing large-scale text corpora is a core challenge in machine learning, crucial for tasks like identifying undesirable model behaviors or biases in training data. Current methods often rely on costly LLM-based techniques (e.g. annotating dataset differences) or dense embedding models (e.g. for clustering), which lack control over the properties of interest. We propose using sparse autoencoders (SAEs) to create SAE embeddings: representations whose dimensions map to interpretable concepts. Through four data analysis tasks, we show that SAE embeddings are more cost-effective and reliable than LLMs and more controllable than dense embeddings. Using the large hypothesis space of SAEs, we can uncover insights such as (1) semantic differences between datasets and (2) unexpected concept correlations in documents. For instance, by comparing model responses, we find that Grok-4 clarifies ambiguities more often than nine other frontier models. Relative to LLMs, SAE embeddings uncover bigger differences at 2-8x lower cost and identify biases more reliably. Additionally, SAE embeddings are controllable: by filtering concepts, we can (3) cluster documents along axes of interest and (4) outperform dense embeddings on property-based retrieval. Using SAE embeddings, we study model behavior with two case studies: investigating how OpenAI model behavior has changed over time and finding "trigger" phrases learned by Tulu-3 (Lambert et al., 2024) from its training data. These results position SAEs as a versatile tool for unstructured data analysis and highlight the neglected importance of interpreting models through their data.