Automatically Interpreting Millions of Features in Large Language Models
Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel pith:3MYT5KLZrecord.jsonopen to challenge →
read the original abstract
While the activations of neurons in deep neural networks usually do not have a simple human-understandable interpretation, sparse autoencoders (SAEs) can be used to transform these activations into a higher-dimensional latent space which may be more easily interpretable. However, these SAEs can have millions of distinct latent features, making it infeasible for humans to manually interpret each one. In this work, we build an open-source automated pipeline to generate and evaluate natural language explanations for SAE features using LLMs. We test our framework on SAEs of varying sizes, activation functions, and losses, trained on two different open-weight LLMs. We introduce five new techniques to score the quality of explanations that are cheaper to run than the previous state of the art. One of these techniques, intervention scoring, evaluates the interpretability of the effects of intervening on a feature, which we find explains features that are not recalled by existing methods. We propose guidelines for generating better explanations that remain valid for a broader set of activating contexts, and discuss pitfalls with existing scoring techniques. We use our explanations to measure the semantic similarity of independently trained SAEs, and find that SAEs trained on nearby layers of the residual stream are highly similar. Our large-scale analysis confirms that SAE latents are indeed much more interpretable than neurons, even when neurons are sparsified using top-$k$ postprocessing. Our code is available at https://github.com/EleutherAI/sae-auto-interp, and our explanations are available at https://huggingface.co/datasets/EleutherAI/auto_interp_explanations.
This paper has not been read by Pith yet.
Forward citations
Cited by 20 Pith papers
-
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.
-
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.
-
Do Sparse Autoencoders Learn Meaningful Concept Hierarchies?
Sparse autoencoders provide a basis for sensible concept hierarchies on visual data but are undermined by hard and soft feature absorption.
-
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.
-
ICA Lens: Interpreting Language Models Without Training Another Dictionary
ICALens applies an optimized ICA workflow to LLM activations and recovers compact interpretable directions that match or exceed public SAEs on SAEBench probing and perturbation tasks without per-layer dictionary training.
-
Interpreting and Steering a Text-to-Speech Language Model with Sparse Autoencoders
Sparse autoencoders on a TTS language model yield interpretable features that causally control attributes such as laughter, gender, and speech rate via targeted interventions.
-
Are Sparse Autoencoder Benchmarks Reliable?
An audit of SAEBench reveals that Targeted Probe Perturbation and Spurious Correlation Removal metrics fail reliability tests and should not be used to evaluate sparse autoencoders.
-
The Rate-Distortion-Polysemanticity Tradeoff in SAEs
SAEs exhibit a rate-distortion-polysemanticity tradeoff where monosemanticity increases rate and distortion, with optimal polysemanticity set by feature co-occurrence probabilities in the data.
-
Why Retrieval-Augmented Generation Fails: A Graph Perspective
Attribution graphs reveal that RAG failures arise from shallow fragmented evidence flow in LLMs, enabling topology-based detection and targeted interventions that reinforce question-guided routing.
-
Stories in Space: In-Context Learning Trajectories in Conceptual Belief Space
LLMs perform in-context learning as trajectories through a structured low-dimensional conceptual belief space, with the structure visible in both behavior and internal representations and causally manipulable via inte...
-
Domain Restriction via Multi SAE Layer Transitions
Multi-layer SAE transitions capture domain-specific signatures that distinguish OOD texts in Gemma-2 models.
-
Tree SAE: Learning Hierarchical Feature Structures in Sparse Autoencoders
Tree SAE learns hierarchical feature pairs in sparse autoencoders by combining activation coverage with a new reconstruction condition, outperforming prior methods on hierarchy detection while remaining competitive on...
-
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 p...
-
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.
-
Dictionary-Aligned Concept Control for Safeguarding Multimodal LLMs
DACO curates a 15,000-concept dictionary from 400K image-caption pairs and uses it to initialize an SAE that enables granular, concept-specific steering of MLLM activations, raising safety scores on MM-SafetyBench and...
-
Making Interpretable Discoveries from Unstructured Data: A High-Dimensional Multiple Hypothesis Testing Approach
A new framework combines AI-derived concept embeddings with high-dimensional selective inference to enable statistically principled, interpretable discovery from unstructured data in empirical economics.
-
Anatomy of Post-Training: Using Interpretability to Characterize Data and Shape the Learning Signal
A new pipeline uses interpretability to characterize concepts in preference data and shape rewards via feature or data interventions during LM post-training.
-
SAEExplainer: Interpreting SAE Features with Activation-Guided Preference Optimization
SAEExplainer applies activation-guided preference optimization in two iterative rounds to improve explanations of SAE features and reduce hallucinations.
-
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.
-
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.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.