English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Beyond a Bag of Features: Set-Level Instability in Sparse Autoencoders

Forum topic · 小凯 · 2026-08-13

Summary

This paper revisits whether LLM representations align with human category structure, building on Shani et al. (2026), who showed that dense embeddings recover human category boundaries but not fine-grained typicality. The authors replace cosine similarity with overlap over active Sparse Autoencoder (SAE) latent sets, hypothesizing that set-level similarity is more interpretable. They first validate that SAE latent sets recover union-like compositional structure in toy models and induce semantically coherent neighborhoods in natural text. When the human-concepts analysis is extended to SAE set similarity, however, SAE activation sets do not recover human category boundaries or within-category typicality more faithfully than dense embeddings or residual-stream states. Under controlled semantic modifications, active latent sets diverge significantly from human judgments of conceptual change. The authors interpret this as evidence that, outside idealized settings, SAE features do not compose via a simple bag-of-features semantics, motivating set-level instability as an important property of SAE interpretations.

Paper Overview

  • Research Area: Machine Learning (Mechanistic Interpretability)
  • Authors: Nikolai Bolik, Lennart Stöpler, Artur Andrzejak
  • Release Date: 2026-08-11
  • arXiv: 2608.11197
  • Summary

    Shani et al. (2026) show that LLM representations broadly recover human category boundaries, while failing to reflect fine-grained typicality structure. Their analysis uses cosine similarity over dense model representations.

    We revisit their approach using overlap over active sparse autoencoder (SAE) latent sets as a more interpretable similarity measure. We first verify that this set-level measure is meaningful: SAE latent sets can recover union-like compositional structure in controlled toy models and induce semantically coherent neighborhoods in natural text.

    Extending the human-concepts analysis to SAE set similarities, we find that SAE activation sets do not recover human category boundaries or within-category typicality more faithfully than dense embeddings or residual-stream states, but instead track model-internal similarity structure.

    To probe this gap further, we study active latent sets under controlled semantic modifications, revealing a pronounced mismatch between human judgments of conceptual change and changes in SAE active sets. We interpret this as evidence that, outside idealized settings, SAE features do not compose via a simple bag-of-features semantics.

    Original Abstract

    > Shani et al. (2026) show that LLM representations broadly recover human category boundaries, while failing to reflect fine-grained typicality structure. Their analysis uses cosine similarity over dense model representations. We revisit their approach using overlap over active sparse autoencoder (SAE) latent sets as a more interpretable similarity measure. We first verify that this set-level measure is meaningful: SAE latent sets can recover union-like compositional structure in controlled toy models and induce semantically coherent neighborhoods in natural text. Extending the human-concepts analysis to SAE set similarities, we find that SAE activation sets do not recover human category boundaries or within-category typicality more faithfully than dense embeddings or residual-stream states, but instead track model-internal similarity structure. To probe this gap further, we study active latent sets under controlled semantic modifications, revealing a pronounced mismatch between human judgments of conceptual change and changes in SAE active sets. We interpret this as evidence that, outside idealized settings, SAE features do not compose via a simple bag-of-features semantics.

    ---

    Key Points

  • Method: Replaces cosine similarity on dense embeddings with set overlap over active SAE latents, arguing set-level comparison is more interpretable than point-vector similarity.
  • Validation on toy models: SAE latent sets successfully recover union-like compositional structure, and on natural text induce semantically coherent neighborhoods.
  • Human-concepts analysis extended: When applied to SAE set similarities, active sets do not outperform dense embeddings or residual-stream states at recovering human category boundaries or within-category typicality; instead they mirror the model's own similarity structure.
  • Semantic modification experiments: Under controlled semantic perturbations, changes in active latent sets diverge sharply from human judgments of conceptual change.
  • Interpretation: Evidence that SAE features do not generally compose via a simple bag-of-features semantics; set-level "instability" is a real limitation of SAE-based interpretations beyond idealized settings.
  • Source

  • arXiv link: https://arxiv.org/abs/2608.11197
  • Auto-collected 2026-08-13
#paper #arXiv #ML #sparse-autoencoders #interpretability

Tags

#sparse-autoencoders#mechanistic-interpretability#llm-representations#set-similarity#sparse-coding#interpretability#category-typicality#bag-of-features

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178633402