Do Sparse Autoencoders Identify Reasoning Features in Language Models?

• URL: http://arxiv.org/abs/2601.05679v3
• Date: Fri, 16 Jan 2026 16:27:07 GMT
• Title: Do Sparse Autoencoders Identify Reasoning Features in Language Models?
• Authors: George Ma, Zhongyuan Liang, Irene Y. Chen, Somayeh Sojoudi,
• Abstract summary: We investigate whether sparse autoencoders (SAEs) identify genuine reasoning features in large language models (LLMs)<n>We first show through a simple theoretical analysis that $ell_$-regularized SAEs are intrinsically biased toward low-dimensional patterns.<n>Motivated by this bias, we introduce a falsification-oriented evaluation framework to test whether feature activation reflects reasoning processes or superficial linguistic correlates.
• Score: 12.693974363520423
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We investigate whether sparse autoencoders (SAEs) identify genuine reasoning features in large language models (LLMs). We first show through a simple theoretical analysis that $\ell_1$-regularized SAEs are intrinsically biased toward low-dimensional patterns, providing a mechanistic explanation for why shallow linguistic cues may be preferentially captured over distributed reasoning behaviors. Motivated by this bias, we introduce a falsification-oriented evaluation framework that combines causal token injection and LLM-guided falsification to test whether feature activation reflects reasoning processes or superficial linguistic correlates. Across 20 configurations spanning multiple model families, layers, and reasoning datasets, we find that features identified by contrastive methods are highly sensitive to token-level interventions, with 45% to 90% activating when a small number of associated tokens are injected into non-reasoning text. For the remaining features, LLM-guided falsification consistently produces non-reasoning inputs that activate the feature and reasoning inputs that do not, with no analyzed feature satisfying our criteria for genuine reasoning behavior. Steering these features yields no improvements in benchmark performance. Overall, our results suggest that SAE features identified by current contrastive approaches primarily capture linguistic correlates of reasoning rather than the underlying reasoning computations themselves. Code is available at https://github.com/GeorgeMLP/reasoning-probing.

Related papers

• Step-Level Sparse Autoencoder for Reasoning Process Interpretation [48.99201531966593]
  Large Language Models (LLMs) have achieved strong complex reasoning capabilities through Chain-of-Thought (CoT) reasoning.<n>We propose step-level sparse autoencoder (SSAE), which serves as an analytical tool to disentangle different aspects of LLMs' reasoning steps into sparse features.<n> Experiments on multiple base models and reasoning tasks show the effectiveness of the extracted features.
  arXiv  Detail & Related papers  (2026-03-03T14:25:02Z)
• Uncovering Hidden Correctness in LLM Causal Reasoning via Symbolic Verification [56.51953062869371]
  DoVerifier is a symbolic verifier that checks whether causal expressions are derivable from a given causal graph using rules from do-calculus and probability theory.<n>Our evaluations on synthetic data and causal QA benchmarks show that DoVerifier more accurately captures semantic correctness of causal reasoning traces.
  arXiv  Detail & Related papers  (2026-01-29T03:22:58Z)
• Reasoning Beyond Chain-of-Thought: A Latent Computational Mode in Large Language Models [39.5490415037017]
  Chain-of-Thought (CoT) prompting has improved the reasoning performance of large language models (LLMs)<n>It remains unclear why it works and whether it is the unique mechanism for triggering reasoning in large language models.
  arXiv  Detail & Related papers  (2026-01-12T23:01:21Z)
• Are Language Models Efficient Reasoners? A Perspective from Logic Programming [109.47572890883248]
  Modern language models (LMs) exhibit strong deductive reasoning capabilities, yet standard evaluations emphasize correctness while overlooking a key aspect of human-like reasoning: efficiency.<n>We propose a framework for assessing LM reasoning efficiency through the lens of logic programming.
  arXiv  Detail & Related papers  (2025-10-29T15:30:31Z)
• Verifying Large Language Models' Reasoning Paths via Correlation Matrix Rank [71.09032766271493]
  Large language models (LLMs) are prone to errors and hallucinations.<n>How to check their outputs effectively and efficiently has become a critical problem in their applications.
  arXiv  Detail & Related papers  (2025-10-28T11:01:10Z)
• Framework for Machine Evaluation of Reasoning Completeness in Large Language Models For Classification Tasks [0.0]
  This paper introduces RACE-Reasoning Alignment for Completeness of Explanations.<n>We analyze four widely used text classification datasets-WIKI ONTOLOGY, AG NEWS, IMDB, and GOEMOTIONS.<n>We show that correct predictions exhibit higher coverage of supporting features, while incorrect predictions are associated with elevated coverage of contradicting features.
  arXiv  Detail & Related papers  (2025-10-23T20:22:22Z)
• Causal Prompting for Implicit Sentiment Analysis with Large Language Models [21.39152516811571]
  Implicit Sentiment Analysis (ISA) aims to infer sentiment that is implied rather than explicitly stated.<n>Recent prompting-based methods using Large Language Models (LLMs) have shown promise in ISA.<n>We propose CAPITAL, a causal prompting framework that incorporates front-door adjustment into CoT reasoning.
  arXiv  Detail & Related papers  (2025-07-01T03:01:09Z)
• Decomposing MLP Activations into Interpretable Features via Semi-Nonnegative Matrix Factorization [17.101290138120564]
  Current methods rely on dictionary learning with sparse autoencoders (SAEs)<n>Here, we tackle these limitations by directly decomposing activations with semi-nonnegative matrix factorization (SNMF)<n>Experiments on Llama 3.1, Gemma 2 and GPT-2 show that SNMF derived features outperform SAEs and a strong supervised baseline (difference-in-means) on causal steering.
  arXiv  Detail & Related papers  (2025-06-12T17:33:29Z)
• Chain-of-Code Collapse: Reasoning Failures in LLMs via Adversarial Prompting in Code Generation [0.3495246564946556]
  Large Language Models (LLMs) have achieved remarkable success in tasks requiring complex reasoning.<n>Do these models truly reason, or do they merely exploit shallow statistical patterns?<n>We introduce Chain-of-Code Collapse, where we investigate the robustness of reasoning LLMs by introducing a suite of semantically faithful yet adversarially structured prompt perturbations.
  arXiv  Detail & Related papers  (2025-06-08T02:43:46Z)
• I Have Covered All the Bases Here: Interpreting Reasoning Features in Large Language Models via Sparse Autoencoders [8.1201445044499]
  Internal mechanisms behind reasoning in LLMs remain unexplored.<n>We employ Sparse Autoencoders to test our hypothesis.<n>Our work provides the first step towards a mechanistic understanding of reasoning in LLMs.
  arXiv  Detail & Related papers  (2025-03-24T16:54:26Z)
• Don't Take Things Out of Context: Attention Intervention for Enhancing Chain-of-Thought Reasoning in Large Language Models [32.71672086718058]
  Few-shot Chain-of-Thought (CoT) significantly enhances the reasoning capabilities of large language models (LLMs)<n>We observe that isolated segments, words, or tokens within CoT demonstrations can unexpectedly disrupt the generation process of LLMs.<n>We propose a Few-shot Attention Intervention method (FAI) that dynamically analyzes the attention patterns of demonstrations to accurately identify these tokens.
  arXiv  Detail & Related papers  (2025-03-14T07:46:33Z)
• Evaluating Human Alignment and Model Faithfulness of LLM Rationale [66.75309523854476]
  We study how well large language models (LLMs) explain their generations through rationales. We show that prompting-based methods are less "faithful" than attribution-based explanations.
  arXiv  Detail & Related papers  (2024-06-28T20:06:30Z)
• CLOMO: Counterfactual Logical Modification with Large Language Models [109.60793869938534]
  We introduce a novel task, Counterfactual Logical Modification (CLOMO), and a high-quality human-annotated benchmark. In this task, LLMs must adeptly alter a given argumentative text to uphold a predetermined logical relationship. We propose an innovative evaluation metric, the Self-Evaluation Score (SES), to directly evaluate the natural language output of LLMs.
  arXiv  Detail & Related papers  (2023-11-29T08:29:54Z)

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.