Workflow
妙笔系统
icon
Search documents
阅文(马宇峰):AIGC在阅文集团的实践落地
Sou Hu Cai Jing· 2025-06-05 16:00
Core Insights - The article discusses the integration of artificial intelligence (AI) in the digital content industry, particularly focusing on the practices of Yuewen Group in the realm of online literature, emphasizing AI as an auxiliary tool for creators rather than a replacement for human creativity [1][3]. Group 1: Intelligent Creation - Yuewen positions intelligent creation as a "comprehensive assistant" for writers, utilizing its self-developed "Miaobi" system to provide robust support across multiple dimensions [1]. - The system enhances the "web literature feel" by training a 200 billion parameter model specifically for web literature genres, improving text processing capabilities from 2K to 32K tokens [1]. - The "Miaobi" system assists in the entire creative process, including world-building, character development, expression optimization, and knowledge consulting [1]. Group 2: Role Playing - Yuewen aims to "revive" characters from stories, enabling deep interaction between readers and characters through data-driven definitions of character essence [1]. - The company employs Retrieval-Augmented Generation (RAG) technology to address issues like hallucinations and outdated knowledge in character dialogues [1]. - By combining supervised fine-tuning and reinforcement learning from human feedback, Yuewen seeks to provide personalized character interactions that surpass generic dialogue models [1]. Group 3: Challenges and Solutions - Yuewen's practices offer reusable solutions to three major challenges in vertical AIGC: defining fields, data storage, and scaling applications [1]. - The company quantifies subjective concepts like "web literature feel" using data and addresses data storage challenges by reducing manual annotation time by 70% through large model-assisted labeling [1]. - To achieve scalable implementation, Yuewen focuses on creating a business closed-loop and optimizing reasoning costs while monitoring core metrics [1][3]. Group 4: Future Outlook - Yuewen's exploration confirms its core belief that AI serves as "auxiliary driving" in creation, with the true control remaining in the hands of writers [3]. - The evolution of AIGC is expected to lead to three major transformations: democratization of creation, interactive reading experiences, and industry upgrades for faster IP incubation [3][4][5]. - The company has accumulated over 17,000 practical cases, establishing a sustainable ecosystem for web literature vertical AIGC through its four-ring model of pre-training, fine-tuning, feedback alignment, and production returns [5].