@inproceedings{chai-etal-2026-activation,
title = "Activation Reward Models for Few-Shot Model Alignment",
author = "Chai, Tianning and
Mitra, Chancharik and
Huang, Brandon and
Gare, Gautam Rajendrakumar and
Lin, Zhiqiu and
Arbelle, Assaf and
Karlinsky, Leonid and
Feris, Rogerio and
Darrell, Trevor and
Ramanan, Deva and
Herzig, Roei",
editor = "Liakata, Maria and
Moreira, Viviane P. and
Zhang, Jiajun and
Jurgens, David",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-acl.1709/",
pages = "34201--34217",
ISBN = "979-8-89176-395-1",
abstract = "Aligning Large Language Models (LLMs) and Large Multimodal Models (LMMs) to human preferences is crucial for improving their real-world behavior. A common approach is to use reward models that enable reinforcement-learning post-training. However, traditional reward modeling requires finetuning on large preference datasets, limiting adaptability to new preferences. We introduce Activation Reward Models (Activation RMs){---}the first mechanistic interpretability approach that steers LLM activations to align with few-shot preference data without finetuning. Our method combines activation denoising and output token likelihood scoring, achieving state-of-the-art performance on standard reward modeling benchmarks, surpassing zero-shot, few-shot, and voting-based baselines. We further demonstrate that Activation RMs mitigate reward hacking behaviors and remain robust to noisy exemplars and spurious reward signals. To evaluate this, we propose PreferenceHack, a novel few-shot benchmark testing reward models on reward hacking in a paired preference format, where Activation RMs achieve state-of-the-art performance, surpassing GPT-4o."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="chai-etal-2026-activation">
<titleInfo>
<title>Activation Reward Models for Few-Shot Model Alignment</title>
</titleInfo>
<name type="personal">
<namePart type="given">Tianning</namePart>
<namePart type="family">Chai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chancharik</namePart>
<namePart type="family">Mitra</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Brandon</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gautam</namePart>
<namePart type="given">Rajendrakumar</namePart>
<namePart type="family">Gare</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhiqiu</namePart>
<namePart type="family">Lin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Assaf</namePart>
<namePart type="family">Arbelle</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Leonid</namePart>
<namePart type="family">Karlinsky</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rogerio</namePart>
<namePart type="family">Feris</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Trevor</namePart>
<namePart type="family">Darrell</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Deva</namePart>
<namePart type="family">Ramanan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Roei</namePart>
<namePart type="family">Herzig</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Maria</namePart>
<namePart type="family">Liakata</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Viviane</namePart>
<namePart type="given">P</namePart>
<namePart type="family">Moreira</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiajun</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Jurgens</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-395-1</identifier>
</relatedItem>
<abstract>Aligning Large Language Models (LLMs) and Large Multimodal Models (LMMs) to human preferences is crucial for improving their real-world behavior. A common approach is to use reward models that enable reinforcement-learning post-training. However, traditional reward modeling requires finetuning on large preference datasets, limiting adaptability to new preferences. We introduce Activation Reward Models (Activation RMs)—the first mechanistic interpretability approach that steers LLM activations to align with few-shot preference data without finetuning. Our method combines activation denoising and output token likelihood scoring, achieving state-of-the-art performance on standard reward modeling benchmarks, surpassing zero-shot, few-shot, and voting-based baselines. We further demonstrate that Activation RMs mitigate reward hacking behaviors and remain robust to noisy exemplars and spurious reward signals. To evaluate this, we propose PreferenceHack, a novel few-shot benchmark testing reward models on reward hacking in a paired preference format, where Activation RMs achieve state-of-the-art performance, surpassing GPT-4o.</abstract>
<identifier type="citekey">chai-etal-2026-activation</identifier>
<location>
<url>https://aclanthology.org/2026.findings-acl.1709/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>34201</start>
<end>34217</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Activation Reward Models for Few-Shot Model Alignment
%A Chai, Tianning
%A Mitra, Chancharik
%A Huang, Brandon
%A Gare, Gautam Rajendrakumar
%A Lin, Zhiqiu
%A Arbelle, Assaf
%A Karlinsky, Leonid
%A Feris, Rogerio
%A Darrell, Trevor
%A Ramanan, Deva
%A Herzig, Roei
%Y Liakata, Maria
%Y Moreira, Viviane P.
%Y Zhang, Jiajun
%Y Jurgens, David
%S Findings of the Association for Computational Linguistics: ACL 2026
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-395-1
%F chai-etal-2026-activation
%X Aligning Large Language Models (LLMs) and Large Multimodal Models (LMMs) to human preferences is crucial for improving their real-world behavior. A common approach is to use reward models that enable reinforcement-learning post-training. However, traditional reward modeling requires finetuning on large preference datasets, limiting adaptability to new preferences. We introduce Activation Reward Models (Activation RMs)—the first mechanistic interpretability approach that steers LLM activations to align with few-shot preference data without finetuning. Our method combines activation denoising and output token likelihood scoring, achieving state-of-the-art performance on standard reward modeling benchmarks, surpassing zero-shot, few-shot, and voting-based baselines. We further demonstrate that Activation RMs mitigate reward hacking behaviors and remain robust to noisy exemplars and spurious reward signals. To evaluate this, we propose PreferenceHack, a novel few-shot benchmark testing reward models on reward hacking in a paired preference format, where Activation RMs achieve state-of-the-art performance, surpassing GPT-4o.
%U https://aclanthology.org/2026.findings-acl.1709/
%P 34201-34217
Markdown (Informal)
[Activation Reward Models for Few-Shot Model Alignment](https://aclanthology.org/2026.findings-acl.1709/) (Chai et al., Findings 2026)
ACL
- Tianning Chai, Chancharik Mitra, Brandon Huang, Gautam Rajendrakumar Gare, Zhiqiu Lin, Assaf Arbelle, Leonid Karlinsky, Rogerio Feris, Trevor Darrell, Deva Ramanan, and Roei Herzig. 2026. Activation Reward Models for Few-Shot Model Alignment. In Findings of the Association for Computational Linguistics: ACL 2026, pages 34201–34217, San Diego, California, United States. Association for Computational Linguistics.