Causality $\neq$ Invariance: Function and Concept Vectors in LLMs

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Opiełka, Gustaw, Rosenbusch, Hannes, Stevenson, Claire E.
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866918357146533888
author Opiełka, Gustaw
Rosenbusch, Hannes
Stevenson, Claire E.
author_facet Opiełka, Gustaw
Rosenbusch, Hannes
Stevenson, Claire E.
contents Do large language models (LLMs) represent concepts abstractly, i.e., independent of input format? We revisit Function Vectors (FVs), compact representations of in-context learning (ICL) tasks that causally drive task performance. Across multiple LLMs, we show that FVs are not fully invariant: FVs are nearly orthogonal when extracted from different input formats (e.g., open-ended vs. multiple-choice), even if both target the same concept. We identify Concept Vectors (CVs), which carry more stable concept representations. Like FVs, CVs are composed of attention head outputs; however, unlike FVs, the constituent heads are selected using Representational Similarity Analysis (RSA) based on whether they encode concepts consistently across input formats. While these heads emerge in similar layers to FV-related heads, the two sets are largely distinct, suggesting different underlying mechanisms. Steering experiments reveal that FVs excel in-distribution, when extraction and application formats match (e.g., both open-ended in English), while CVs generalize better out-of-distribution across both question types (open-ended vs. multiple-choice) and languages. Our results show that LLMs do contain abstract concept representations, but these differ from those that drive ICL performance.
format Preprint
id arxiv_https___arxiv_org_abs_2602_22424
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Causality $\neq$ Invariance: Function and Concept Vectors in LLMs
Opiełka, Gustaw
Rosenbusch, Hannes
Stevenson, Claire E.
Computation and Language
Machine Learning
Do large language models (LLMs) represent concepts abstractly, i.e., independent of input format? We revisit Function Vectors (FVs), compact representations of in-context learning (ICL) tasks that causally drive task performance. Across multiple LLMs, we show that FVs are not fully invariant: FVs are nearly orthogonal when extracted from different input formats (e.g., open-ended vs. multiple-choice), even if both target the same concept. We identify Concept Vectors (CVs), which carry more stable concept representations. Like FVs, CVs are composed of attention head outputs; however, unlike FVs, the constituent heads are selected using Representational Similarity Analysis (RSA) based on whether they encode concepts consistently across input formats. While these heads emerge in similar layers to FV-related heads, the two sets are largely distinct, suggesting different underlying mechanisms. Steering experiments reveal that FVs excel in-distribution, when extraction and application formats match (e.g., both open-ended in English), while CVs generalize better out-of-distribution across both question types (open-ended vs. multiple-choice) and languages. Our results show that LLMs do contain abstract concept representations, but these differ from those that drive ICL performance.
title Causality $\neq$ Invariance: Function and Concept Vectors in LLMs
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2602.22424