Pith. sign in

REVIEW 1 cited by

The Knowledge Microscope: Features as Better Analytical Lenses than Neurons

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2502.12483 v2 pith:FMP4DFKG submitted 2025-02-18 cs.CL

The Knowledge Microscope: Features as Better Analytical Lenses than Neurons

classification cs.CL
keywords featuresneuronsknowledgeanalyticalbetterexpressioninterpretabilityunits
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Previous studies primarily utilize MLP neurons as units of analysis for understanding the mechanisms of factual knowledge in Language Models (LMs); however, neurons suffer from polysemanticity, leading to limited knowledge expression and poor interpretability. In this paper, we first conduct preliminary experiments to validate that Sparse Autoencoders (SAE) can effectively decompose neurons into features, which serve as alternative analytical units. With this established, our core findings reveal three key advantages of features over neurons: (1) Features exhibit stronger influence on knowledge expression and superior interpretability. (2) Features demonstrate enhanced monosemanticity, showing distinct activation patterns between related and unrelated facts. (3) Features achieve better privacy protection than neurons, demonstrated through our proposed FeatureEdit method, which significantly outperforms existing neuron-based approaches in erasing privacy-sensitive information from LMs.Code and dataset will be available.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Towards Atoms of Large Language Models

    cs.CL 2025-09 reject novelty 4.0

    The authors define 'atoms' as sparse, near-orthogonal directions in LLM representations under a data-adaptive inner product, and show threshold-activated sparse autoencoders can recover them with about 99.9% reconstru...