Metraipolis › archive › dc_aec487ad545e

Beyond a Bag of Features: Set-Level Instability in Sparse Autoencoders — arXiv:2608.11197

Source: arXiv (en)
Original: https://arxiv.org/abs/2608.11197
License (as archived): CC0-1.0
Archived: 2026-10-19T06:14:34.644Z
SHA-256 of the archived text: 1c08d15458092aed146e1db7db1e5cd9e7c697e80947862d5b8650097e7f2cf2
Stable URL of this page: https://metraipolis.com/documento/dc_aec487ad545e

How to cite

Beyond a Bag of Features: Set-Level Instability in Sparse Autoencoders — arXiv:2608.11197 — arXiv (en), https://arxiv.org/abs/2608.11197 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_aec487ad545e (Metraipolis archive dc_aec487ad545e, sha256 1c08d15458092aed146e1db7db1e5cd9e7c697e80947862d5b8650097e7f2cf2)

BibTeX

@misc{dcaec487ad545e,
  title = {Beyond a Bag of Features: Set-Level Instability in Sparse Autoencoders — arXiv:2608.11197},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.11197},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_aec487ad545e; sha256 1c08d15458092aed146e1db7db1e5cd9e7c697e80947862d5b8650097e7f2cf2},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.11197. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

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 well-controlled semantic modifications, revealing a substantial mismatch between human judgements of conceptual change and change in the SAE active set. We interpret this as evidence that, outside idealised settings, SAE features do not compose via simple bag-of-features semantics.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-12 in cs.CL).
Categorie arXiv: Machine Learning (cs.LG) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2608.11197 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-19T06:08:50.376Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-08-12?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).