@inproceedings{kim-etal-2025-drift,
title = "Drift: Decoding-time Personalized Alignments with Implicit User Preferences",
author = "Kim, Minbeom and
Lee, Kang-il and
Joo, Seongho and
Lee, Hwaran and
Thonet, Thibaut and
Jung, Kyomin",
editor = "Christodoulopoulos, Christos and
Chakraborty, Tanmoy and
Rose, Carolyn and
Peng, Violet",
booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2025",
month = nov,
year = "2025",
address = "Suzhou, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.findings-emnlp.324/",
pages = "6107--6126",
ISBN = "979-8-89176-335-7",
abstract = "Personalized alignments towards individual users have been a long-standing goal in large language models (LLMs). We introduce Drift, a novel framework that personalizes LLMs at decoding time with implicit user preferences. Unlike traditional Reinforcement Learning from Human Feedback (RLHF), which relies on vast annotated datasets and expensive gradient updates, Drift operates in a training-free manner by steering a frozen LLM through few-shot preference modeling. Our approach represents user preferences as a composition of interpretable and predefined attributes, and employs a zero-shot rewarding mechanism based on contrastive system prompts. Experiments on both a synthetic persona dataset Perspective and a real human-annotated dataset PRISM demonstrate that Drift achieves performance comparable to standard RLHF methods while using only 50{--}100 examples. Our results show that Drift delivers not only computationally efficient but also interpretable personalization."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kim-etal-2025-drift">
<titleInfo>
<title>Drift: Decoding-time Personalized Alignments with Implicit User Preferences</title>
</titleInfo>
<name type="personal">
<namePart type="given">Minbeom</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kang-il</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Seongho</namePart>
<namePart type="family">Joo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hwaran</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Thibaut</namePart>
<namePart type="family">Thonet</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kyomin</namePart>
<namePart type="family">Jung</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2025</title>
</titleInfo>
<name type="personal">
<namePart type="given">Christos</namePart>
<namePart type="family">Christodoulopoulos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tanmoy</namePart>
<namePart type="family">Chakraborty</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Carolyn</namePart>
<namePart type="family">Rose</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Violet</namePart>
<namePart type="family">Peng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Suzhou, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-335-7</identifier>
</relatedItem>
<abstract>Personalized alignments towards individual users have been a long-standing goal in large language models (LLMs). We introduce Drift, a novel framework that personalizes LLMs at decoding time with implicit user preferences. Unlike traditional Reinforcement Learning from Human Feedback (RLHF), which relies on vast annotated datasets and expensive gradient updates, Drift operates in a training-free manner by steering a frozen LLM through few-shot preference modeling. Our approach represents user preferences as a composition of interpretable and predefined attributes, and employs a zero-shot rewarding mechanism based on contrastive system prompts. Experiments on both a synthetic persona dataset Perspective and a real human-annotated dataset PRISM demonstrate that Drift achieves performance comparable to standard RLHF methods while using only 50–100 examples. Our results show that Drift delivers not only computationally efficient but also interpretable personalization.</abstract>
<identifier type="citekey">kim-etal-2025-drift</identifier>
<location>
<url>https://aclanthology.org/2025.findings-emnlp.324/</url>
</location>
<part>
<date>2025-11</date>
<extent unit="page">
<start>6107</start>
<end>6126</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Drift: Decoding-time Personalized Alignments with Implicit User Preferences
%A Kim, Minbeom
%A Lee, Kang-il
%A Joo, Seongho
%A Lee, Hwaran
%A Thonet, Thibaut
%A Jung, Kyomin
%Y Christodoulopoulos, Christos
%Y Chakraborty, Tanmoy
%Y Rose, Carolyn
%Y Peng, Violet
%S Findings of the Association for Computational Linguistics: EMNLP 2025
%D 2025
%8 November
%I Association for Computational Linguistics
%C Suzhou, China
%@ 979-8-89176-335-7
%F kim-etal-2025-drift
%X Personalized alignments towards individual users have been a long-standing goal in large language models (LLMs). We introduce Drift, a novel framework that personalizes LLMs at decoding time with implicit user preferences. Unlike traditional Reinforcement Learning from Human Feedback (RLHF), which relies on vast annotated datasets and expensive gradient updates, Drift operates in a training-free manner by steering a frozen LLM through few-shot preference modeling. Our approach represents user preferences as a composition of interpretable and predefined attributes, and employs a zero-shot rewarding mechanism based on contrastive system prompts. Experiments on both a synthetic persona dataset Perspective and a real human-annotated dataset PRISM demonstrate that Drift achieves performance comparable to standard RLHF methods while using only 50–100 examples. Our results show that Drift delivers not only computationally efficient but also interpretable personalization.
%U https://aclanthology.org/2025.findings-emnlp.324/
%P 6107-6126
Markdown (Informal)
[Drift: Decoding-time Personalized Alignments with Implicit User Preferences](https://aclanthology.org/2025.findings-emnlp.324/) (Kim et al., Findings 2025)
ACL