Enregistré dans:
Détails bibliographiques
Auteurs principaux: Firooz, Hamed, Liu, Rui, Lu, Yuchen, Hou, Zhenyu, Xiong, Fangzhou, Zhang, Xiaoyang, Jian, Changshu, Zhu, Zhicheng, Ma, Jiayuan, Tao, Jacob, Gupta, Chaitali, Peng, Xiaochang, Mei, Shike, Cui, Hang, Qin, Yang, Tang, Shuo, Gaedtke, Jason, Mittal, Arpit
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:https://arxiv.org/abs/2512.20061
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866909974530097152
author Firooz, Hamed
Liu, Rui
Lu, Yuchen
Hou, Zhenyu
Xiong, Fangzhou
Zhang, Xiaoyang
Jian, Changshu
Zhu, Zhicheng
Ma, Jiayuan
Tao, Jacob
Gupta, Chaitali
Peng, Xiaochang
Mei, Shike
Cui, Hang
Qin, Yang
Tang, Shuo
Gaedtke, Jason
Mittal, Arpit
author_facet Firooz, Hamed
Liu, Rui
Lu, Yuchen
Hou, Zhenyu
Xiong, Fangzhou
Zhang, Xiaoyang
Jian, Changshu
Zhu, Zhicheng
Ma, Jiayuan
Tao, Jacob
Gupta, Chaitali
Peng, Xiaochang
Mei, Shike
Cui, Hang
Qin, Yang
Tang, Shuo
Gaedtke, Jason
Mittal, Arpit
contents Content moderation at scale remains one of the most pressing challenges in today's digital ecosystem, where billions of user- and AI-generated artifacts must be continuously evaluated for policy violations. Although recent advances in large language models (LLMs) have demonstrated strong potential for policy-grounded moderation, the practical challenges of training these systems to achieve expert-level accuracy in real-world settings remain largely unexplored, particularly in regimes characterized by label sparsity, evolving policy definitions, and the need for nuanced reasoning beyond shallow pattern matching. In this work, we present a comprehensive empirical investigation of scaling reinforcement learning (RL) for content classification, systematically evaluating multiple RL training recipes and reward-shaping strategies-including verifiable rewards and LLM-as-judge frameworks-to transform general-purpose language models into specialized, policy-aligned classifiers across three real-world content moderation tasks. Our findings provide actionable insights for industrial-scale moderation systems, demonstrating that RL exhibits sigmoid-like scaling behavior in which performance improves smoothly with increased training data, rollouts, and optimization steps before gradually saturating. Moreover, we show that RL substantially improves performance on tasks requiring complex policy-grounded reasoning while achieving up to 100x higher data efficiency than supervised fine-tuning, making it particularly effective in domains where expert annotations are scarce or costly.
format Preprint
id arxiv_https___arxiv_org_abs_2512_20061
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Scaling Reinforcement Learning for Content Moderation with Large Language Models
Firooz, Hamed
Liu, Rui
Lu, Yuchen
Hou, Zhenyu
Xiong, Fangzhou
Zhang, Xiaoyang
Jian, Changshu
Zhu, Zhicheng
Ma, Jiayuan
Tao, Jacob
Gupta, Chaitali
Peng, Xiaochang
Mei, Shike
Cui, Hang
Qin, Yang
Tang, Shuo
Gaedtke, Jason
Mittal, Arpit
Artificial Intelligence
Content moderation at scale remains one of the most pressing challenges in today's digital ecosystem, where billions of user- and AI-generated artifacts must be continuously evaluated for policy violations. Although recent advances in large language models (LLMs) have demonstrated strong potential for policy-grounded moderation, the practical challenges of training these systems to achieve expert-level accuracy in real-world settings remain largely unexplored, particularly in regimes characterized by label sparsity, evolving policy definitions, and the need for nuanced reasoning beyond shallow pattern matching. In this work, we present a comprehensive empirical investigation of scaling reinforcement learning (RL) for content classification, systematically evaluating multiple RL training recipes and reward-shaping strategies-including verifiable rewards and LLM-as-judge frameworks-to transform general-purpose language models into specialized, policy-aligned classifiers across three real-world content moderation tasks. Our findings provide actionable insights for industrial-scale moderation systems, demonstrating that RL exhibits sigmoid-like scaling behavior in which performance improves smoothly with increased training data, rollouts, and optimization steps before gradually saturating. Moreover, we show that RL substantially improves performance on tasks requiring complex policy-grounded reasoning while achieving up to 100x higher data efficiency than supervised fine-tuning, making it particularly effective in domains where expert annotations are scarce or costly.
title Scaling Reinforcement Learning for Content Moderation with Large Language Models
topic Artificial Intelligence
url https://arxiv.org/abs/2512.20061