aiqbench analysis
This research introduces sparse autoencoders (SAEs) as a cost-effective and controllable alternative to large language models for analyzing text corpora, offering superior reliability in detecting biases and semantic differences. By mapping embedding dimensions directly to interpretable concepts, the toolkit enables precise clustering of documents along specific axes while significantly reducing computational costs compared to LLM-based annotation methods. These findings advance machine learning interpretability by providing a scalable framework for uncovering hidden model behaviors without relying on expensive proprietary inference.
Interpretable Embeddings with Sparse Autoencoders: A Data Analysis Toolkit
arXiv:2512.10092v2 Announce Type: replace 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.
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.
ICML 2026. Project page and code: this https URL
Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2512.10092 [cs.AI]
(or arXiv:2512.10092v2 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2512.10092 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 339 words
Related coverage
PIRAMID: Progress and Plans
LessWrong · Jul 27, 2026
One Lens, Many Worlds : A Capability-Typed Interface for World-Model Interpretability
arXiv · Jun 10, 2026
Interpreting GFlowNets for Drug Discovery: What probes can and cannot show
arXiv · Aug 6, 2026
Bias Analysis of L2 Speaking Assessment Systems Using Concept Activation Vectors
arXiv · Aug 7, 2026