国家预印本平台

This paper examines how generative artificial intelligence (AI) agents trained on digital traces of deceased loved ones reshape remembrance as a distributed, co performed emotional process—a phenomenon we term digital nostalgia. While mainstream discourse focuses on technical fidelity or data privacy, it overlooks the dynamic, dialogical structure of AI mediated Nostalgia. Adopting a conceptual analytical approach that integrates distributed cognition theory and phenomenological accounts of nostalgia, we argue that AI memory agents play a dual role: as cognitive artifacts that store and trigger memories, and as affective quasi others whose responsive presence transforms solitary recollection into collaborative emotional performance. This transformation reconfigures the spatial dimension already implicit in classical accounts of nostalgia: the longed for “place of belonging” is no longer a remembered physical location or an imagined scene, but a relational space enacted in the present through dialogue. Building on this framework, we propose emotional smoothing as a testable conceptual hypothesis: a multi layered mechanism—spanning training data bias, reinforcement learning from human feedback (RLHF), and user experience flow—through which AI’s “appropriate,” empathetic responses may systematically trim trauma, ambiguity, and contradiction from memory narratives. Drawing on recent empirical studies of user interactions with generative ghosts, we suggest how smoothing might extend from emotional regulation to cognitive narrowing, and how nostalgia may become commodified. The paper contributes: (1) a phenomenological framework for conceptualizing AI mediated nostalgia as spatial and dialogical; (2) a critical analysis of emotional smoothing as a hypothesis shaped by commercial incentives; and (3) an application of the thanatosensitivity framework to interrogate how current AI memory agents may risk erasing the ambivalence and complexity of grief. We conclude with implications for thanatosensitive design and policy, arguing that preserving emotional friction—if supported by future empirical research—may be essential for authentic human AI interaction (HAI).

除了恒星这类发光天体之外，还存在一类天体，它们并非由核反应驱动，而是由物质在引力势阱中的吸积提供能量。这类天体包括主序前星、密近双星系统、活动星系核和类星体，以及可能还包括某些类型的超新星和伽马射线暴。尽管与普通恒星相比，这些天体都较为罕见，但它们在许多物理和观测应用中都具有重要的意义。在这些天体中，吸积的物理过程在许多情况下是相似的。人们重点考察的密近双星系统，这也是人们考察最深入的由吸积驱动的天体。但是，之前的学者对这部分内容的描述过于艰深难懂，用到大量数学和物理方程，而我在本文中则采用通俗的方式解释密近双星及其形成的吸积盘。这是为了让普通中学生和天文爱好者也能轻易了解这一重要的天文知识。

《雨旸气候亲机》是宋代道教雷法体系中一部实用性很强的天气预测手册，收录于明《正统道藏》。全书由诗文、云图、占示和笔记等部分构成，专为修持雷法的道士把握施法时机而作。它的核心，是将观云、辨风这类经验技艺，与以北斗星象为核心的占候传统交织在一起。书中所记的“久雨西风又放晴”等谚语、三十九幅形态化云图，以及“白云遮斗，明日大变”一类占辞，都显示出一种独特的认知方式：经验气象学提供当下微候的佐证，占星学则给出根本的气运解释，二者在“天人感应”的框架中互为支撑。对当时的使用者而言，这并非迷信与科学的杂糅，而是一套用以预测风雨、安顿身心，且被认为行之有效的整体知识。

生成式人工智能（Generative Artificial Intelligence，GenAI）在赋能公共图书馆古籍数字化的同时，也带来了数据失真、模型“幻觉”与权责模糊等可信风险，故亟须加强系统性可信治理。本研究首先分析GenAI赋能古籍数字化过程中引发的可信风险，并构建适配古籍特性的涵盖基础层、内容层、知识层与服务层的可信度评估框架；然后，据此构建包含13项评估指标的GenAI赋能古籍数字化可信度评估指标体系，实现对生成内容的结构化、可操作的可信度评估；最后，提出建立分层响应机制、加强资源分级管理与权责协同的制度保障、构建可信治理长效机制三个方面的可信治理策略，形成“可信风险分析—可信度评估—可信治理”的完整闭环，旨在为GenAI在古籍数字化过程中的合规化、高质量的应用提供理论与实践参考。

在生成式人工智能（Generative Artificial Intelligence，GenAI）背景下，对中外个人数据保护法律政策进行量化评价与比较分析，可为我国强化GenAI应用过程中的个人数据风险治理能力提供有益借鉴。本研究利用隐含狄利克雷分布（Latent Dirichlet Allocation，LDA）主题模型对39份法律政策文本进行主题挖掘，结合现有研究成果，设计GenAI背景下个人数据保护法律政策评价体系，并运用政策一致性（PolicyModeling Consistency，PMC）指数模型对其中29份已实施法律政策文本进行量化评价，进而从政策功能定位、实施逻辑、体系完备性方面比较中国、美国、英国以及欧盟的政策差异。基于中外比较结果，本研究从兼顾安全与发展、强化敏捷治理模式、优化多元协同格局三个方面梳理总结对我国的启示，以期为提升我国GenAI背景下的个人数据保护水平提供参考。