Interpretable Embeddings with Sparse Autoencoders: A Data Analysis Toolkit
Authors:
- Nick JiangNJNick Jiang
- Xiaoqing SunXSXiaoqing Sun
- Lewis SmithLSLewis Smith
- Neel NandaNNNeel Nanda
Preprint
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.Keywords
All Related Versions
This publication has 22 references indexed in Scilit:
- LITA: An Efficient LLM-Assisted Iterative Topic Augmentation Framework2025
- DocETL: Agentic Query Rewriting and Evaluation for Complex Document ProcessingProceedings of the VLDB Endowment, 2025
- Large Language Models Enable Few-Shot ClusteringTransactions of the Association for Computational Linguistics, 2024
- The POLAR Framework: Polar Opposites Enable Interpretability of Pre-Trained Word Embeddings2020
- Nuanced Metrics for Measuring Unintended Bias with Real Data for Text Classification2019
- Survey of Word Co-occurrence Measures for Collocation DetectionComputación y Sistemas, 2016
- Interactive topic modelingMachine Learning, 2013
- A similarity measure for indefinite rankingsACM Transactions on Information Systems, 2010
- The Probabilistic Relevance Framework: BM25 and BeyondFoundations and Trends® in Information Retrieval, 2009
- A vector space model for automatic indexingCommunications of the ACM, 1975