Steering LLMs? Actually, Sparse Autoencoders can outperform simple baselines

• URL: http://arxiv.org/abs/2605.31183v1
• Date: Fri, 29 May 2026 11:53:54 GMT
• Title: Steering LLMs? Actually, Sparse Autoencoders can outperform simple baselines
• Abstract summary: This work serves as a partial rebuttal for Sparse Autoencoders and suggests that the results of Wu et al. (2025) did not do them full justice.<n>We find that Sparse Autoencoders can, in fact, perform close to on par with the reference LoRA performance on the AxBench benchmark.<n>We also find that our pipeline selects features that are surprisingly causal of their identified labels when using only its interpretability-based components.
• Score: 3.652509571098291
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Sparse Autoencoders (SAEs) have been seen as a promising avenue for exploring the internals of Large Language Models (LLMs) and for steering model output generation. When AxBench - a model steering benchmark - was introduced in Wu et al. (2025), SAEs did not seem to live up to their original hype due to poor steering performance relative to a set of simple baselines. This work serves as a partial rebuttal for Sparse Autoencoders and suggests that the results of Wu et al. (2025) did not do them full justice. We find that Sparse Autoencoders can, in fact, perform close to on par with the reference LoRA performance on the AxBench benchmark, when features are selected and labelled with our supervised pipeline. We also find that our pipeline selects features that are surprisingly causal of their identified labels when using only its interpretability-based components. Lastly, we present evidence that high sparsity (low l0) may not be crucial for successful steering based on interpretability, which is in contrast to the earlier findings in Wang et al. (2025).

Related papers

• 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)
• Enhancing LLM Steering through Sparse Autoencoder-Based Vector Refinement [31.282134977964976]
  Existing steering methods rely on large-scale datasets to learn clear behavioral information.<n>We introduce Refinement of Steering Vector via Sparse Autoencoder (SAE-RSV) that leverages SAEs to semantically denoise and augment the steering vectors.<n>In our framework, we first remove task-irrelevant features according to their semantics provided by SAEs, and then enrich task-relevant features missing from the small dataset through their semantic similarity to the identified relevant features.
  arXiv  Detail & Related papers  (2025-09-28T10:49:22Z)
• CorrSteer: Generation-Time LLM Steering via Correlated Sparse Autoencoder Features [1.5874067490843806]
  We propose CorrSteer, which selects features by correlating sample correctness with SAE activations from generated tokens at inference time.<n>Our work establishes correlation-based selection as an effective and scalable approach for automated SAE steering across language model applications.
  arXiv  Detail & Related papers  (2025-08-18T00:01:42Z)
• Feature Extraction and Steering for Enhanced Chain-of-Thought Reasoning in Language Models [48.40096116617163]
  Large Language Models (LLMs) demonstrate the ability to solve reasoning and mathematical problems using the Chain-of-Thought (CoT) technique.<n>This work, inspired by the deep thinking paradigm of DeepSeek-R1, utilizes a steering technique to enhance the reasoning ability of an LLM without external datasets.
  arXiv  Detail & Related papers  (2025-05-21T15:17:59Z)
• START: Self-taught Reasoner with Tools [51.38785489790888]
  We introduce START (Self-Taught Reasoner with Tools), a tool-integrated long Chain-of-thought (CoT) reasoning LLM.<n> START is capable of performing complex computations, self-checking, exploring diverse methods, and self-ging.<n>It significantly outperforms the base QwQ-32B and achieves performance comparable to the state-of-the-art open-weight model R1-Distill-Qwen-32B.
  arXiv  Detail & Related papers  (2025-03-06T17:11:51Z)
• 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)
• ACCORD: Closing the Commonsense Measurability Gap [16.572584339052753]
  ACCORD is a framework for disentangling the commonsense grounding and reasoning abilities of large language models (LLMs)<n>It introduces formal elements to commonsense reasoning to explicitly control and quantify reasoning complexity beyond the typical 1 or 2 hops.<n>It can automatically generate benchmarks of arbitrary reasoning complexity, and so it scales with future LLM improvements.
  arXiv  Detail & Related papers  (2024-06-04T22:08:24Z)
• Gait Recognition in the Wild: A Large-scale Benchmark and NAS-based Baseline [95.88825497452716]
  Gait benchmarks empower the research community to train and evaluate high-performance gait recognition systems. GREW is the first large-scale dataset for gait recognition in the wild. SPOSGait is the first NAS-based gait recognition model.
  arXiv  Detail & Related papers  (2022-05-05T14:57:39Z)

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.