Large Language Models (LLMs) have advanced automated writing assistance, enabling complex tasks like co-writing novels and poems. However, real-world writing typically requires various support and collaboration across stages and scenarios. Existing research mainly examines how writers utilize single text generators, neglecting this broader context. This paper introduces Inspo, a web-based editor that incorporates various text generators and online crowd workers. Through a three-phase user study, we examine writers' interactions with Inspo for novel writing. Quantitative analyses of writing logs highlight changes in participants' writing progress and the influence of various text-generation models. Complementing this with qualitative insights from semi-structured interviews, we illustrate participants' perceptions of these models and the crowd. Based on the findings, we provide design recommendations for the next generation of intelligent writing tools and discuss the potential sociocultural implications of integrating AI and human input in the writing process.

翻译：大语言模型（LLMs）推动了自动化写作辅助技术的发展，使得合著小说与诗歌等复杂任务成为可能。然而，现实世界的写作通常需要跨阶段、跨场景的多维度支持与协作。现有研究主要关注作家如何使用单一文本生成器，忽视了这一更广泛的背景。本文介绍了Inspo——一款集成了多种文本生成器与在线众包工作者的网页编辑器。通过三阶段用户研究，我们考察了作家使用Inspo进行小说写作时的交互模式。对写作日志的定量分析揭示了参与者写作进程的变化以及不同文本生成模型的影响。结合半结构化访谈的定性洞察，我们阐述了参与者对这些模型及众包者的认知。基于研究发现，我们为下一代智能写作工具提供了设计建议，并探讨了在写作过程中整合人工智能与人类输入的潜在社会文化影响。