Saved in:
Bibliographic Details
Main Authors: Ma, George, Liang, Zhongyuan, Chen, Irene Y., Sojoudi, Somayeh
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.05679
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909050961133568
author Ma, George
Liang, Zhongyuan
Chen, Irene Y.
Sojoudi, Somayeh
author_facet Ma, George
Liang, Zhongyuan
Chen, Irene Y.
Sojoudi, Somayeh
contents We study how reliably sparse autoencoders (SAEs) support claims about reasoning-related internal features in large language models. We first give a stylized analysis showing that sparsity-regularized decoding can preferentially retain stable low-dimensional correlates while suppressing high-dimensional within-behavior variation, motivating the possibility that contrastively selected "reasoning" features may concentrate on cue-like structure when such cues are coupled with reasoning traces. Building on this perspective, we propose a falsification-based evaluation framework that combines causal token injection with LLM-guided counterexample construction. Across 22 configurations spanning multiple model families, layers, and reasoning datasets, we find that many contrastively selected candidates are highly sensitive to token-level interventions, with 45%-90% activating after injecting only a few associated tokens into non-reasoning text. For the remaining context-dependent candidates, LLM-guided falsification produces targeted non-reasoning inputs that trigger activation and meaning-preserving paraphrases of top-activating reasoning traces that suppress it. A small steering study yields minimal changes on the evaluated benchmarks. Overall, our results suggest that, in the settings we study, sparse decompositions can favor low-dimensional correlates that co-occur with reasoning, underscoring the need for falsification when attributing high-level behaviors to individual SAE features. Code is available at https://github.com/GeorgeMLP/reasoning-probing.
format Preprint
id arxiv_https___arxiv_org_abs_2601_05679
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Do Sparse Autoencoders Identify Reasoning Features in Language Models?
Ma, George
Liang, Zhongyuan
Chen, Irene Y.
Sojoudi, Somayeh
Machine Learning
We study how reliably sparse autoencoders (SAEs) support claims about reasoning-related internal features in large language models. We first give a stylized analysis showing that sparsity-regularized decoding can preferentially retain stable low-dimensional correlates while suppressing high-dimensional within-behavior variation, motivating the possibility that contrastively selected "reasoning" features may concentrate on cue-like structure when such cues are coupled with reasoning traces. Building on this perspective, we propose a falsification-based evaluation framework that combines causal token injection with LLM-guided counterexample construction. Across 22 configurations spanning multiple model families, layers, and reasoning datasets, we find that many contrastively selected candidates are highly sensitive to token-level interventions, with 45%-90% activating after injecting only a few associated tokens into non-reasoning text. For the remaining context-dependent candidates, LLM-guided falsification produces targeted non-reasoning inputs that trigger activation and meaning-preserving paraphrases of top-activating reasoning traces that suppress it. A small steering study yields minimal changes on the evaluated benchmarks. Overall, our results suggest that, in the settings we study, sparse decompositions can favor low-dimensional correlates that co-occur with reasoning, underscoring the need for falsification when attributing high-level behaviors to individual SAE features. Code is available at https://github.com/GeorgeMLP/reasoning-probing.
title Do Sparse Autoencoders Identify Reasoning Features in Language Models?
topic Machine Learning
url https://arxiv.org/abs/2601.05679