@inproceedings{cho-etal-2026-features,
title = "Features in Motion: Periodicity, Homeostasis, and Adaptivity in {LLM} Few-Shot Prompting",
author = "Cho, Ikhyun and
Lee, Sewoong and
Hockenmaier, Julia",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.1878/",
pages = "42819--42838",
ISBN = "979-8-89176-459-0",
abstract = "In this work, we identify three representation-level dynamics shared across diverse large language models (LLMs) during the processing of few-shot prompts. These structural behaviors are not explicitly trained, yet they emerge reliably across model families and scales to systematically shape internal representations. Using Sparse Autoencoders (SAEs) as an analytical lens, we uncover three primary phenomena: (1) a decimal periodicity effect, where specific features exhibit a repeating modulo-10 activation pattern; (2) local feature homeostasis, a compensatory push{--}pull dynamic in which current representations are inversely modulated by prior exemplars; and (3) adaptive feature redistribution, wherein specific features flexibly shift their semantic profiles in response to contextual constraints. We validate these mechanisms across multiple model families and prompt formats. Together, they offer fresh insights into the internal dynamics of in-context learning and raise important caveats for future SAE-based interpretability studies."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="cho-etal-2026-features">
<titleInfo>
<title>Features in Motion: Periodicity, Homeostasis, and Adaptivity in LLM Few-Shot Prompting</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ikhyun</namePart>
<namePart type="family">Cho</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sewoong</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Julia</namePart>
<namePart type="family">Hockenmaier</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>In this work, we identify three representation-level dynamics shared across diverse large language models (LLMs) during the processing of few-shot prompts. These structural behaviors are not explicitly trained, yet they emerge reliably across model families and scales to systematically shape internal representations. Using Sparse Autoencoders (SAEs) as an analytical lens, we uncover three primary phenomena: (1) a decimal periodicity effect, where specific features exhibit a repeating modulo-10 activation pattern; (2) local feature homeostasis, a compensatory push–pull dynamic in which current representations are inversely modulated by prior exemplars; and (3) adaptive feature redistribution, wherein specific features flexibly shift their semantic profiles in response to contextual constraints. We validate these mechanisms across multiple model families and prompt formats. Together, they offer fresh insights into the internal dynamics of in-context learning and raise important caveats for future SAE-based interpretability studies.</abstract>
<identifier type="citekey">cho-etal-2026-features</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1878/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>42819</start>
<end>42838</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Features in Motion: Periodicity, Homeostasis, and Adaptivity in LLM Few-Shot Prompting
%A Cho, Ikhyun
%A Lee, Sewoong
%A Hockenmaier, Julia
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F cho-etal-2026-features
%X In this work, we identify three representation-level dynamics shared across diverse large language models (LLMs) during the processing of few-shot prompts. These structural behaviors are not explicitly trained, yet they emerge reliably across model families and scales to systematically shape internal representations. Using Sparse Autoencoders (SAEs) as an analytical lens, we uncover three primary phenomena: (1) a decimal periodicity effect, where specific features exhibit a repeating modulo-10 activation pattern; (2) local feature homeostasis, a compensatory push–pull dynamic in which current representations are inversely modulated by prior exemplars; and (3) adaptive feature redistribution, wherein specific features flexibly shift their semantic profiles in response to contextual constraints. We validate these mechanisms across multiple model families and prompt formats. Together, they offer fresh insights into the internal dynamics of in-context learning and raise important caveats for future SAE-based interpretability studies.
%U https://aclanthology.org/2026.emnlp-main.1878/
%P 42819-42838
Markdown (Informal)
[Features in Motion: Periodicity, Homeostasis, and Adaptivity in LLM Few-Shot Prompting](https://aclanthology.org/2026.emnlp-main.1878/) (Cho et al., EMNLP 2026)
ACL