Quantifying Feature Space Universality Across Large Language Models via Sparse Autoencoders

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Lan, Michael, Torr, Philip, Meek, Austin, Khakzar, Ashkan, Krueger, David, Barez, Fazl
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913849190383616
author Lan, Michael
Torr, Philip
Meek, Austin
Khakzar, Ashkan
Krueger, David
Barez, Fazl
author_facet Lan, Michael
Torr, Philip
Meek, Austin
Khakzar, Ashkan
Krueger, David
Barez, Fazl
contents The Universality Hypothesis in large language models (LLMs) claims that different models converge towards similar concept representations in their latent spaces. Providing evidence for this hypothesis would enable researchers to exploit universal properties, facilitating the generalization of mechanistic interpretability techniques across models. Previous works studied if LLMs learned the same features, which are internal representations that activate on specific concepts. Since comparing features across LLMs is challenging due to polysemanticity, in which LLM neurons often correspond to multiple unrelated features rather than to distinct concepts, sparse autoencoders (SAEs) have been employed to disentangle LLM neurons into SAE features corresponding to distinct concepts. In this paper, we introduce a new variation of the universality hypothesis called Analogous Feature Universality: we hypothesize that even if SAEs across different models learn different feature representations, the spaces spanned by SAE features are similar, such that one SAE space is similar to another SAE space under rotation-invariant transformations. Evidence for this hypothesis would imply that interpretability techniques related to latent spaces, such as steering vectors, may be transferred across models via certain transformations. To investigate this hypothesis, we first pair SAE features across different models via activation correlation, and then measure spatial relation similarities between paired features via representational similarity measures, which transform spaces into representations that reveal hidden relational similarities. Our experiments demonstrate high similarities for SAE feature spaces across various LLMs, providing evidence for feature space universality.
format Preprint
id arxiv_https___arxiv_org_abs_2410_06981
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Quantifying Feature Space Universality Across Large Language Models via Sparse Autoencoders
Lan, Michael
Torr, Philip
Meek, Austin
Khakzar, Ashkan
Krueger, David
Barez, Fazl
Machine Learning
Artificial Intelligence
Computation and Language
The Universality Hypothesis in large language models (LLMs) claims that different models converge towards similar concept representations in their latent spaces. Providing evidence for this hypothesis would enable researchers to exploit universal properties, facilitating the generalization of mechanistic interpretability techniques across models. Previous works studied if LLMs learned the same features, which are internal representations that activate on specific concepts. Since comparing features across LLMs is challenging due to polysemanticity, in which LLM neurons often correspond to multiple unrelated features rather than to distinct concepts, sparse autoencoders (SAEs) have been employed to disentangle LLM neurons into SAE features corresponding to distinct concepts. In this paper, we introduce a new variation of the universality hypothesis called Analogous Feature Universality: we hypothesize that even if SAEs across different models learn different feature representations, the spaces spanned by SAE features are similar, such that one SAE space is similar to another SAE space under rotation-invariant transformations. Evidence for this hypothesis would imply that interpretability techniques related to latent spaces, such as steering vectors, may be transferred across models via certain transformations. To investigate this hypothesis, we first pair SAE features across different models via activation correlation, and then measure spatial relation similarities between paired features via representational similarity measures, which transform spaces into representations that reveal hidden relational similarities. Our experiments demonstrate high similarities for SAE feature spaces across various LLMs, providing evidence for feature space universality.
title Quantifying Feature Space Universality Across Large Language Models via Sparse Autoencoders
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2410.06981