Are Sparse Autoencoder Benchmarks Reliable?

• URL: http://arxiv.org/abs/2605.18229v1
• Date: Mon, 18 May 2026 11:20:57 GMT
• Title: Are Sparse Autoencoder Benchmarks Reliable?
• Abstract summary: We audit the SAE quality metrics in SAEBench, the de-facto standard SAE evaluation suite.<n>We find that two metrics, Targeted Probe Perturbation and Spurious Correlation Removal, fail multiple lenses at their canonical settings.<n>The sae-probes variant of $k$-sparse probing is the most reliable metric we tested, but even sae-probes struggles to separate variants of the same SAE architecture.
• Score: 2.0789144738685956
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Sparse autoencoders (SAEs) are a core interpretability tool for large language models, and progress on SAE architectures depends on benchmarks that reliably distinguish better SAEs from worse ones. We audit the SAE quality metrics in SAEBench, the de-facto standard SAE evaluation suite, through three complementary lenses: reseed noise on a fixed SAE, ground-truth correlation on synthetic SAEs, and discriminability across training trajectories. We find that two of these metrics, Targeted Probe Perturbation (TPP) and Spurious Correlation Removal (SCR), fail multiple lenses at their canonical settings and should not be used to evaluate SAEs. The other metrics show higher reseed noise and lower discriminability than the field assumes. The sae-probes variant of $k$-sparse probing is the most reliable metric we tested, but even sae-probes struggles to separate variants of the same SAE architecture. Our results show the field needs better SAE benchmarks.

Related papers

• From Sparse Features to Trustworthy Proxies: Certifying SAE-Based Interpretability [53.681678236115836]
  We study when can an SAE-based explanation be treated as a faithful view of an underlying frozen LM.<n>Our framework derives an upper bound on the base model's expected risk using four measurable quantities.<n>We show that the bound becomes non-vacuous on GPT-2 Small, Gemma-2B, and Llama-3-8B at practical sample sizes.
  arXiv  Detail & Related papers  (2026-06-16T18:28:23Z)
• SynthSAEBench: Evaluating Sparse Autoencoders on Scalable Realistic Synthetic Data [7.429838797007225]
  We introduce SynthSAEBench, a toolkit for generating large-scale synthetic data with realistic feature characteristics.<n>Our benchmark reproduces several previously observed LLM SAE phenomena, including the disconnect between reconstruction and latent quality metrics.<n>We also use our benchmark to identify a new failure mode: Matching Pursuit SAEs exploit superposition noise to improve reconstruction without learning ground-truth features.
  arXiv  Detail & Related papers  (2026-02-16T12:22:00Z)
• Sanity Checks for Sparse Autoencoders: Do SAEs Beat Random Baselines? [10.871959954490217]
  Sparse Autoencoders (SAEs) have emerged as a promising tool for interpreting neural networks by decomposing their activations into sparse sets of human-interpretable features.<n>Recent work has introduced multiple SAE variants and successfully scaled them to frontier models.<n>Despite much excitement, a growing number of negative results in downstream tasks casts doubt on whether SAEs recover meaningful features.
  arXiv  Detail & Related papers  (2026-02-15T11:53:55Z)
• Does higher interpretability imply better utility? A Pairwise Analysis on Sparse Autoencoders [63.544453925182005]
  We train 90 SAEs across three language models and evaluate their interpretability and steering utility.<n>Our analysis reveals only a relatively weak positive association (tau b approx 0.298), indicating that interpretability is an insufficient proxy for steering performance.<n>We propose a novel selection criterion called Delta Token Confidence, which measures how much amplifying a feature changes the next token distribution.
  arXiv  Detail & Related papers  (2025-10-04T04:14:50Z)
• SVeritas: Benchmark for Robust Speaker Verification under Diverse Conditions [54.34001921326444]
  Speaker verification (SV) models are increasingly integrated into security, personalization, and access control systems.<n>Existing benchmarks evaluate only subsets of these conditions, missing others entirely.<n>We introduce SVeritas, a comprehensive Speaker Verification tasks benchmark suite, assessing SV systems under stressors like recording duration, spontaneity, content, noise, microphone distance, reverberation, channel mismatches, audio bandwidth, codecs, speaker age, and susceptibility to spoofing and adversarial attacks.
  arXiv  Detail & Related papers  (2025-09-21T14:11:16Z)
• AHELM: A Holistic Evaluation of Audio-Language Models [78.20477815156484]
  multimodal audio-language models (ALMs) take interleaved audio and text as input and output text.<n>AHELM is a benchmark that aggregates various datasets -- including 2 new synthetic audio-text datasets called PARADE and CoRe-Bench.<n>We also standardize the prompts, inference parameters, and evaluation metrics to ensure equitable comparisons across models.
  arXiv  Detail & Related papers  (2025-08-29T07:40:39Z)
• Ensembling Sparse Autoencoders [10.81463830315253]
  Sparse autoencoders (SAEs) are used to decompose neural network activations into human-interpretable features.<n>We propose to ensemble multiple SAEs through naive bagging and boosting.<n>Our empirical results demonstrate that ensembling SAEs can improve the reconstruction of language model activations, diversity of features, and SAE stability.
  arXiv  Detail & Related papers  (2025-05-21T23:31:21Z)
• SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability [2.4557811577651436]
  SAEBench is a comprehensive evaluation suite that measures SAE performance across eight diverse metrics.<n>We open-source a suite of over 200 SAEs across eight recently proposed SAE architectures and training algorithms.<n>Our evaluation reveals that gains on proxy metrics do not reliably translate to better practical performance.
  arXiv  Detail & Related papers  (2025-03-12T16:49:02Z)
• Are Sparse Autoencoders Useful? A Case Study in Sparse Probing [6.836374436707495]
  Sparse autoencoders (SAEs) are a popular method for interpreting concepts represented in large language model (LLM) activations.<n>One alternative source of evidence would be demonstrating that SAEs improve performance on downstream tasks beyond existing baselines.<n>We test this by applying SAEs to the real-world task of LLM activation probing in four regimes.
  arXiv  Detail & Related papers  (2025-02-23T18:54:15Z)
• AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoders [73.37603699731329]
  We introduce AxBench, a large-scale benchmark for steering and concept detection.<n>For steering, we find that prompting outperforms all existing methods, followed by finetuning.<n>For concept detection, representation-based methods such as difference-in-means, perform the best.
  arXiv  Detail & Related papers  (2025-01-28T18:51:24Z)
• Llama Scope: Extracting Millions of Features from Llama-3.1-8B with Sparse Autoencoders [115.34050914216665]
  Sparse Autoencoders (SAEs) have emerged as a powerful unsupervised method for extracting sparse representations from language models. We introduce a suite of 256 SAEs, trained on each layer and sublayer of the Llama-3.1-8B-Base model, with 32K and 128K features. We assess the generalizability of SAEs trained on base models to longer contexts and fine-tuned models.
  arXiv  Detail & Related papers  (2024-10-27T17:33:49Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.