# Sparse Autoencoders for Hypothesis Generation

arXiv: https://arxiv.org/abs/2502.04382 · HTML: https://mlsanity.com/paper/2502.04382/

## Key facts

- Cited by 26 papers in the map (data through 14 Oct 2026).
- Placed in SAE features and their applications, one of 13 directions in Mechanistic interpretability.
- Builds on 21 of its 55 references that are in the map.
- First posted on arXiv 5 Feb 2025.

## Placed in

- Mechanistic interpretability / [SAE features and their applications](https://mlsanity.com/mechinterp/sae-features-and-their-applications/)

## Builds on (most cited first, 21 of 21)

- [Pointer Sentinel Mixture Models](https://mlsanity.com/paper/1609.07843/) (2016)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)](https://mlsanity.com/paper/1711.11279/) (2017)
- [Concept Bottleneck Models](https://mlsanity.com/paper/2007.04612/) (2020)
- [BERTopic: Neural topic modeling with a class-based TF-IDF procedure](https://mlsanity.com/paper/2203.05794/) (2022)
- [Scaling and evaluating sparse autoencoders](https://mlsanity.com/paper/2406.04093/) (2024)
- [Can Large Language Models Transform Computational Social Science?](https://mlsanity.com/paper/2305.03514/) (2023)
- [k-Sparse Autoencoders](https://mlsanity.com/paper/1312.5663/) (2013)
- [BERTology Meets Biology: Interpreting Attention in Protein Language Models](https://mlsanity.com/paper/2006.15222/) (2020)
- [Concept Induction: Analyzing Unstructured Text with High-Level Concepts Using LLooM](https://mlsanity.com/paper/2404.12259/) (2024)
- [Hypothesis Generation with Large Language Models](https://mlsanity.com/paper/2404.04326/) (2024)
- [Analyzing Polarization in Social Media: Method and Application to Tweets on 21 Mass Shootings](https://mlsanity.com/paper/1904.01596/) (2019)
- [Concept Bottleneck Large Language Models](https://mlsanity.com/paper/2412.07992/) (2024)
- [Disentangling Dense Embeddings with Sparse Autoencoders](https://mlsanity.com/paper/2408.00657/) (2024)
- [Interpretable-by-Design Text Understanding with Iteratively Generated Concept Bottleneck](https://mlsanity.com/paper/2310.19660/) (2023)
- [Explaining Datasets in Words: Statistical Models with Natural Language Parameters](https://mlsanity.com/paper/2409.08466/) (2024)
- [Bayesian Concept Bottleneck Models with LLM Priors](https://mlsanity.com/paper/2410.15555/) (2024)
- [Are Neural Topic Models Broken?](https://mlsanity.com/paper/2210.16162/) (2022)
- [Using large language models to promote health equity](https://mlsanity.com/paper/2312.14804/) (2023)
- [A large language model-based approach to quantifying the effects of social determinants in liver transplant decisions](https://mlsanity.com/paper/2412.07924/) (2024)
- [Clinical Notes Reveal Physician Fatigue](https://mlsanity.com/paper/2312.03077/) (2023)

## Built on by (most cited first, 26 of 26)

- [Correlated Errors in Large Language Models](https://mlsanity.com/paper/2506.07962/) (2025)
- [Steering Out-of-Distribution Generalization with Concept Ablation Fine-Tuning](https://mlsanity.com/paper/2507.16795/) (2025)
- [HypoBench: Towards Systematic and Principled Benchmarking for Hypothesis Generation](https://mlsanity.com/paper/2504.11524/) (2025)
- [What's In My Human Feedback? Learning Interpretable Descriptions of Preference Data](https://mlsanity.com/paper/2510.26202/) (2025)
- [Position: Use Sparse Autoencoders to Discover Unknowns](https://mlsanity.com/paper/2506.23845/) (2025)
- [How Many Features Can a Language Model Store Under the Linear Representation Hypothesis?](https://mlsanity.com/paper/2602.11246/) (2026)
- [Model Directions, Not Words: Mechanistic Topic Models Using Sparse Autoencoders](https://mlsanity.com/paper/2507.23220/) (2025)
- [From Reasoning to Learning: A Survey on Hypothesis Discovery and Rule Learning with Large Language Models](https://mlsanity.com/paper/2505.21935/) (2025)
- [Measuring Sparse Autoencoder Feature Sensitivity](https://mlsanity.com/paper/2509.23717/) (2025)
- [Exploratory Causal Inference in SAEnce](https://mlsanity.com/paper/2510.14073/) (2025)
- [Ensembling Sparse Autoencoders](https://mlsanity.com/paper/2505.16077/) (2025)
- [Sparse Autoencoders, Again?](https://mlsanity.com/paper/2506.04859/) (2025)
- [Accelerating Social Science Research via Agentic Hypothesization and Experimentation](https://mlsanity.com/paper/2602.07983/) (2026)
- [Can SAEs reveal and mitigate racial biases of LLMs in healthcare?](https://mlsanity.com/paper/2511.00177/) (2025)
- [ComplLLM: Fine-tuning LLMs to Discover Complementary Signals for Decision-making](https://mlsanity.com/paper/2602.19458/) (2026)
- [Position: Behavioural Assurance Cannot Verify the Safety Claims Governance Now Demands](https://mlsanity.com/paper/2605.15164/) (2026)
- [Interestingness First Classifiers](https://mlsanity.com/paper/2508.19780/) (2025)
- [Data-Centric Interpretability for LLM-based Multi-Agent Reinforcement Learning](https://mlsanity.com/paper/2602.05183/) (2026)
- [Disentangling Steering Vectors](https://mlsanity.com/paper/2609.07037/) (2026)
- [Steered Generation via Gradient-Based Optimization on Sparse Query Features](https://mlsanity.com/paper/2605.23040/) (2026)
- [Position: Prioritize Identifying Structure, Not Complex Models, for Scientific Discovery](https://mlsanity.com/paper/2606.02632/) (2026)
- [PairSAE: Mechanistic Interpretability from Pair Representations in Protein Co-Folding](https://mlsanity.com/paper/2606.27440/) (2026)
- [What Makes Something Hard(er)? Explaining Question Difficulty in Natural Language](https://mlsanity.com/paper/2610.01627/) (2026)
- [Clinical Concept Centers in LLMs](https://mlsanity.com/paper/2610.02829/) (2026)
- [A Testable Theory of Atomic Features](https://mlsanity.com/paper/2610.05794/) (2026)
- [RouterInterp: Understanding Superposed Specialisation in Mixture of Experts Routing](https://mlsanity.com/paper/2610.11775/) (2026)

## Abstract (arXiv)

We describe HypotheSAEs, a general method to hypothesize interpretable relationships between text data (e.g., headlines) and a target variable (e.g., clicks). HypotheSAEs has three steps: (1) train a sparse autoencoder on text embeddings to produce interpretable features describing the data distribution, (2) select features that predict the target variable, and (3) generate a natural language interpretation of each feature (e.g., "mentions being surprised or shocked") using an LLM. Each interpretation serves as a hypothesis about what predicts the target variable. Compared to baselines, our method better identifies reference hypotheses on synthetic datasets (at least +0.06 in F1) and produces more predictive hypotheses on real datasets (~twice as many significant findings), despite requiring 1-2 orders of magnitude less compute than recent LLM-based methods. HypotheSAEs also produces novel discoveries on two well-studied tasks: explaining partisan differences in Congressional speeches and identifying drivers of engagement with online headlines.