Many distinct SAE features share identical explanations, with the average annotation resolving only 70% of feature identity in a large annotated dataset.
hub Canonical reference
Scaling and evaluating sparse autoencoders
Canonical reference. 73% of citing Pith papers cite this work as background.
51
Pith papers citing it
Background
73% of classified citations
abstract
Sparse autoencoders provide a promising unsupervised approach for extracting interpretable features from a language model by reconstructing activations from a sparse bottleneck layer. Since language models learn many concepts, autoencoders need to be very large to recover all relevant features. However, studying the properties of autoencoder scaling is difficult due to the need to balance reconstruction and sparsity objectives and the presence of dead latents. We propose using k-sparse autoencoders [Makhzani and Frey, 2013] to directly control sparsity, simplifying tuning and improving the reconstruction-sparsity frontier. Additionally, we find modifications that result in few dead latents, even at the largest scales we tried. Using these techniques, we find clean scaling laws with respect to autoencoder size and sparsity. We also introduce several new metrics for evaluating feature quality based on the recovery of hypothesized features, the explainability of activation patterns, and the sparsity of downstream effects. These metrics all generally improve with autoencoder size. To demonstrate the scalability of our approach, we train a 16 million latent autoencoder on GPT-4 activations for 40 billion tokens. We release training code and autoencoders for open-source models, as well as a visualizer.
hub tools
citation-role summary
background 8
method 3
citation-polarity summary
representative citing papers
WriteSAE introduces sparse autoencoders with rank-1 matrix atoms for recurrent state updates, allowing replacement tests that outperform deletion on 92.4% of positions and a formula predicting logit changes with R²=0.98.
SLAM achieves 100% detection on Gemma-2 models with only 1-2 point quality cost by causally steering SAE-identified residual-stream directions for linguistic structure.
SASA replaces single-vector decoders in SAEs with learned subspaces plus block sparsity and nuclear-norm regularization, proving that a single group becomes the global minimizer once block size meets intrinsic dimension and yielding polynomial rather than exponential sample complexity.
CEDAR learns an invertible rotation of vision-language embeddings to concentrate semantics into sparse, axis-aligned coordinates for improved interpretability.
GAE reduces the faithfulness gap in dictionary-based explainers under distribution shift by geometrically realigning the ID dictionary to the OOD-active subspace, with a quadratic excess-loss bound.
Foundation models yield less human-interpretable features than supervised vision transformers, with interpretability tied to activation locality and coarse semantic alignment rather than task performance.
SAEParate disentangles sparse representations in diffusion models via contrastive clustering and nonlinear encoding to enable more precise concept unlearning with reduced side effects.
Optimal INR freeze depth matches highest weight stable rank layer; SAEs reveal SIREN atoms are localized while FFMLP atoms trace cohort contours with causal impact on PSNR.
Linear readouts incur an Omega(d^{-1/2}) crosstalk floor that caps the Hanni template at d^{3/2} capacity, while threshold recovery succeeds at quadratic loads for s = O(d/log d) sparsity, resolving the apparent contradiction via distinct readout invariants.
Sparse autoencoders on ViT class tokens reveal stable Class Activation Profiles for in-distribution data, enabling OOD detection via divergence from core energy profiles.
Cross-Layer Transcoders decompose ViT activations into sparse, depth-aware layer contributions that maintain zero-shot accuracy and enable faithful attribution of the final representation.
Sparse autoencoders applied to a 14.5M-parameter clinical EHR model reveal progressive abstraction across layers, with SAE features outperforming dense ones for mortality in full-sequence probes but not in leakage-safe windows where dense representations match or exceed them.
The paper proposes information scope as a new interpretability axis for SAE features in CLIP and introduces the Contextual Dependency Score to separate local from global scope features, showing they influence model predictions differently.
Joint training of a primary SAE with a meta SAE that applies a decomposability penalty on decoder directions produces more atomic latents, shown by 7.5% lower mean absolute phi and 7.6% higher fuzzing scores on GPT-2.
A new SAE-based framework extracts visual, textual, and multimodal concepts from VLMs and reports up to 45% better visual concept quality on a VQA dataset while identifying multimodal concepts.
RefMem-Bench benchmarks reflective memory in dialogue with 26K instances across eight dimensions, and REMIND improves model accuracy via hierarchical evidence retrieval, grounding, and abstraction.
Transcoders decompose MLP layers in Gemma 3-4B-IT to trace visual grounding more effectively than SAEs and predict hallucinations from circuit graph features at AUC 0.68.
SAE-FT uses a sparse autoencoder on pre-trained CLIP visual representations to regularize fine-tuning by penalizing changes to semantically meaningful features, aiming for robust performance on ImageNet and distribution shifts.
A latent mediation framework with sparse autoencoders enables non-additive token-level influence attribution in LLMs by learning orthogonal features and back-propagating attributions.
DECO is a sparse MoE architecture with ReLU-based routing, learnable expert scaling, and NormSiLU activation that matches dense Transformer performance at 20% expert activation and delivers 2.93x speedup on Jetson AGX Orin.
Steering Dark Triad features in an LLM increases exploitative and aggressive behavior while leaving strategic deception and cognitive empathy unchanged, indicating dissociable antisocial pathways.
Causal dimensionality kappa of transformer layers grows sub-linearly with SAE width, remains invariant to model scale, and stays constant across depth while attribution thresholds drop sharply.
Tree SAE learns hierarchical feature structures by combining activation coverage with a new reconstruction condition, outperforming prior SAEs on hierarchical pair detection while matching state-of-the-art benchmark performance.
citing papers explorer
-
Descriptive Collision in Sparse Autoencoder Auto-Interpretability: When One Explanation Describes Many Features
Many distinct SAE features share identical explanations, with the average annotation resolving only 70% of feature identity in a large annotated dataset.
-
WriteSAE: Sparse Autoencoders for Recurrent State
WriteSAE introduces sparse autoencoders with rank-1 matrix atoms for recurrent state updates, allowing replacement tests that outperform deletion on 92.4% of positions and a formula predicting logit changes with R²=0.98.
-
SLAM: Structural Linguistic Activation Marking for Language Models
SLAM achieves 100% detection on Gemma-2 models with only 1-2 point quality cost by causally steering SAE-identified residual-stream directions for linguistic structure.
-
Subspace-Aware Sparse Autoencoders for Effective Mechanistic Interpretability
SASA replaces single-vector decoders in SAEs with learned subspaces plus block sparsity and nuclear-norm regularization, proving that a single group becomes the global minimizer once block size meets intrinsic dimension and yielding polynomial rather than exponential sample complexity.
-
Conceptualizing Embeddings: Sparse Disentanglement for Vision-Language Models
CEDAR learns an invertible rotation of vision-language embeddings to concentrate semantics into sparse, axis-aligned coordinates for improved interpretability.
-
Geometry-Adaptive Explainer for Faithful Dictionary-Based Interpretability under Distribution Shift
GAE reduces the faithfulness gap in dictionary-based explainers under distribution shift by geometrically realigning the ID dictionary to the OOD-active subspace, with a quadratic excess-loss bound.
-
Capability $\neq$ Interpretability: Human Interpretability of Vision Foundation Models
Foundation models yield less human-interpretable features than supervised vision transformers, with interpretability tied to activation locality and coarse semantic alignment rather than task performance.
-
Disentangled Sparse Representations for Concept-Separated Diffusion Unlearning
SAEParate disentangles sparse representations in diffusion models via contrastive clustering and nonlinear encoding to enable more precise concept unlearning with reduced side effects.
-
What Cohort INRs Encode and Where to Freeze Them
Optimal INR freeze depth matches highest weight stable rank layer; SAEs reveal SIREN atoms are localized while FFMLP atoms trace cohort contours with causal impact on PSNR.
-
Linear-Readout Floors and Threshold Recovery in Computation in Superposition
Linear readouts incur an Omega(d^{-1/2}) crosstalk floor that caps the Hanni template at d^{3/2} capacity, while threshold recovery succeeds at quadratic loads for s = O(d/log d) sparsity, resolving the apparent contradiction via distinct readout invariants.
-
Sparsity as a Key: Unlocking New Insights from Latent Structures for Out-of-Distribution Detection
Sparse autoencoders on ViT class tokens reveal stable Class Activation Profiles for in-distribution data, enabling OOD detection via divergence from core energy profiles.
-
Can Cross-Layer Transcoders Replace Vision Transformer Activations? An Interpretable Perspective on Vision
Cross-Layer Transcoders decompose ViT activations into sparse, depth-aware layer contributions that maintain zero-shot accuracy and enable faithful attribution of the final representation.
-
Sparse Autoencoder Decomposition of Clinical Sequence Model Representations: Feature Complexity, Task Specialisation, and Mortality Prediction
Sparse autoencoders applied to a 14.5M-parameter clinical EHR model reveal progressive abstraction across layers, with SAE features outperforming dense ones for mortality in full-sequence probes but not in leakage-safe windows where dense representations match or exceed them.
-
Beyond Semantics: Disentangling Information Scope in Sparse Autoencoders for CLIP
The paper proposes information scope as a new interpretability axis for SAE features in CLIP and introduces the Contextual Dependency Score to separate local from global scope features, showing they influence model predictions differently.
-
MetaSAEs: Joint Training with a Decomposability Penalty Produces More Atomic Sparse Autoencoder Latents
Joint training of a primary SAE with a meta SAE that applies a decomposability penalty on decoder directions produces more atomic latents, shown by 7.5% lower mean absolute phi and 7.6% higher fuzzing scores on GPT-2.
-
Extraction and Analysis of Multimodal Concepts in Vision Language Models through Sparse Autoencoders
A new SAE-based framework extracts visual, textual, and multimodal concepts from VLMs and reports up to 45% better visual concept quality on a VQA dataset while identifying multimodal concepts.
-
Connecting the Dots: Benchmarking Reflective Memory in Long-Horizon Dialogue
RefMem-Bench benchmarks reflective memory in dialogue with 26K instances across eight dimensions, and REMIND improves model accuracy via hierarchical evidence retrieval, grounding, and abstraction.
-
Transcoders Trace Visual Grounding and Hallucinations in Vision-Language Models
Transcoders decompose MLP layers in Gemma 3-4B-IT to trace visual grounding more effectively than SAEs and predict hallucinations from circuit graph features at AUC 0.68.
-
Sparse Autoencoders enable Robust and Interpretable Fine-tuning of CLIP models
SAE-FT uses a sparse autoencoder on pre-trained CLIP visual representations to regularize fine-tuning by penalizing changes to semantically meaningful features, aiming for robust performance on ImageNet and distribution shifts.
-
Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces
A latent mediation framework with sparse autoencoders enables non-additive token-level influence attribution in LLMs by learning orthogonal features and back-propagating attributions.
-
DECO: Sparse Mixture-of-Experts with Dense-Comparable Performance on End-Side Devices
DECO is a sparse MoE architecture with ReLU-based routing, learnable expert scaling, and NormSiLU activation that matches dense Transformer performance at 20% expert activation and delivers 2.93x speedup on Jetson AGX Orin.
-
Exploitation Without Deception: Dark Triad Feature Steering Reveals Separable Antisocial Circuits in Language Models
Steering Dark Triad features in an LLM increases exploitative and aggressive behavior while leaving strategic deception and cognitive empathy unchanged, indicating dissociable antisocial pathways.
-
Causal Dimensionality of Transformer Representations: Measurement, Scaling, and Layer Structure
Causal dimensionality kappa of transformer layers grows sub-linearly with SAE width, remains invariant to model scale, and stays constant across depth while attribution thresholds drop sharply.
-
Tree SAE: Learning Hierarchical Feature Structures in Sparse Autoencoders
Tree SAE learns hierarchical feature structures by combining activation coverage with a new reconstruction condition, outperforming prior SAEs on hierarchical pair detection while matching state-of-the-art benchmark performance.
-
Emergent Symbolic Structure in Health Foundation Models: Extraction, Alignment, and Cross-Modal Transfer
Health foundation model embeddings contain an interpretable symbolic organization shared across modalities that supports cross-domain transfer without joint training.
-
From Token Lists to Graph Motifs: Weisfeiler-Lehman Analysis of Sparse Autoencoder Features
Graph-motif clustering of SAE features via a frequency-binned WL kernel recovers structural families not captured by decoder cosine similarity or token histograms.
-
The Granularity Axis: A Micro-to-Macro Latent Direction for Social Roles in Language Models
LLMs organize prompted social roles along a dominant, stable, and causally steerable granularity axis in representation space that runs from micro to macro levels.
-
Feature Starvation as Geometric Instability in Sparse Autoencoders
Adaptive elastic net SAEs (AEN-SAEs) mitigate feature starvation in SAEs by combining ℓ2 structural stability with adaptive ℓ1 reweighting, producing a Lipschitz-continuous sparse coding map that recovers global feature support under mild assumptions.
-
Pairwise matrices for sparse autoencoders: single-feature inspection mislabels causal axes
Pairwise matrices for SAEs demonstrate that single-feature inspection mislabels causal axes, with joint suppression and matched-geometry controls revealing distinct output regimes not captured by single-feature or random perturbations.
-
GeoSAE: Geometric Prior-Guided Layer-Wise Sparse Autoencoder Annotation of Brain MRI Foundation Models
GeoSAE extracts a compact, interpretable feature set from frozen brain MRI foundation models that predicts MCI-to-AD conversion (AUC 0.746) with age-deconfounded annotations and replicates across cohorts.
-
LatentDiff: Scaling Semantic Dataset Comparison to Millions of Images
LatentDiff scales semantic dataset comparison to millions of images using latent spaces of vision encoders combined with sparse autoencoders and density ratio estimation, showing better accuracy and robustness than caption-based approaches on a new benchmark for sparse distribution shifts.
-
Towards Understanding the Robustness of Sparse Autoencoders
Integrating pretrained sparse autoencoders into LLM residual streams reduces jailbreak success rates by up to 5x across multiple models and attacks.
-
Geometric Routing Enables Causal Expert Control in Mixture of Experts
Cosine-similarity routing in low-dimensional space makes MoE experts monosemantic by construction and enables direct causal control via centroid interventions.
-
Improving Robustness In Sparse Autoencoders via Masked Regularization
Masked regularization in sparse autoencoders disrupts token co-occurrences to reduce feature absorption, enhance probing, and narrow OOD gaps across architectures and sparsity levels.
-
Understanding Emergent Misalignment via Feature Superposition Geometry
Emergent misalignment occurs because fine-tuning amplifies target features that overlap geometrically with harmful ones in superposition, and filtering samples near toxic features mitigates it.
-
Phase-Associative Memory: Sequence Modeling in Complex Hilbert Space
PAM, a complex-valued associative memory model, exhibits steeper power-law scaling in loss and perplexity than a matched real-valued baseline when trained on WikiText-103 from 5M to 100M parameters.
-
Sparse Autoencoders as a Steering Basis for Phase Synchronization in Graph-Based CFD Surrogates
Sparse autoencoders enable phase synchronization in frozen graph CFD surrogates through Hilbert-identified oscillatory features and SVD-based time-varying rotations.
-
In your own words: computationally identifying interpretable themes in free-text survey data
A computational framework identifies more coherent themes in free-text survey data on race, gender, and sexual orientation than previous methods, with applications for survey design, explaining variation, and detecting identity discordance.
-
Graph-Regularized Sparse Autoencoders for LLM Safety Steering
GSAE improves selective refusal on safety benchmarks by smoothing SAE directions over a co-activation graph and applying them via a two-gate controller, outperforming standard SAEs and baselines on Llama-3 and other models.
-
Beyond Cross-Modal Alignment: Measuring and Leveraging Modality Gap in Vision-Language Models
Introduces Modality Dominance Score (MDS) to measure modality-specific features in VLMs and applies training-free editing to improve bias mitigation, adversarial generation, and modality control.
-
Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models
Sparse feature circuits are introduced as interpretable causal subnetworks in language models, supporting unsupervised discovery of thousands of circuits and a method called SHIFT to improve classifier generalization by ablating irrelevant features.
-
Sparse probes and murky physics: a case study of interpretability challenges in a foundation model for continuum dynamics
Case study applies SAE probing with enstrophy triage to a continuum-dynamics foundation model and reports intermittent feature consistency that does not align with standard physics while linking some output discrepancies to specific feature changes.
-
Characterize Then Distill: Mechanistic Reasoning in Large Output Spaces
Reasoning in large output spaces proceeds via shortlisting then fine-grained reasoning; this characterization enables a mechanistic distillation strategy that outperforms standard distillation.
-
Steered Generation via Gradient-Based Optimization on Sparse Query Features
Prototype-Based Sparse Steering decomposes query activations with SAEs and optimizes sparse features via gradients to steer LLM outputs toward specific behaviors.
-
Features have life history. And we should care
Language model features form an early stable carrier scaffold of about 50 sparse features that is load-bearing, predictable from onset firing, and recruits most later features.
-
Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models
The survey organizes mechanistic interpretability techniques into a Locate-Steer-Improve framework to enable actionable improvements in LLM alignment, capability, and efficiency.
-
Safe-SAIL: Towards a Fine-grained Safety Landscape of Large Language Models via Sparse Autoencoder Interpretation Framework
Safe-SAIL supplies a pre-explanation metric and segment-level simulation to interpret 1758 safety SAE features across pornography, politics, violence, and terror, with public models and tools released.
-
At the Edge of Understanding: Sparse Autoencoders Trace The Limits of Transformer Generalization
Sparse autoencoders show OOD prompts increase fallacious concept activation in transformers, offering a mechanistic measure of shift and a path to robust fine-tuning.
- REALISTA: Realistic Latent Adversarial Attacks that Elicit LLM Hallucinations
- From Mechanistic to Compositional Interpretability