摘要： 大语言模型正以前所未有的规模介入知识生产，其影响不止于效率的提升，更在于重构知识被理解、判断与继承的方式。以“互文性”为核心特征的生成机制，使文本在既有话语之间能够频繁流转重组，呈现出高度流畅、逻辑自洽且广泛可接受的知识形态。但在这种看似“合理”的生成过程中，差异、否定与判断正被悄然消解，复杂问题被提前收敛为安全稳定的答案，知识生产由此显现出一种隐蔽且结构性的“平庸之恶”倾向。这种平庸并非错误或低质知识，而是强调互文性的自动化、规模化与去主体性，导致判断责任的转移与主体位置的退场。由此，知识生产越来越依赖算法对既有语料的概率拟合，不再通过争论、质疑与反思展开，只能沿着技术互文性的既定路径被不断确认与强化。若长此以往，知识生产中独有的灵韵、人性、情感、体验将逐渐被湮灭。就此而言，技术进步所带来的真正挑战，是避免知识在持续生成中丧失张力与深度。只有在技术设计中为不确定性和多重理解保留空间，在制度安排中防止效率逻辑对创新路径形成低阶锁定，并在文化层面重新强调判断、责任与内在体悟的价值，知识生产才可能摆脱平庸化的引力，重新通向新问题、新意义与新可能。

关键词: 互文性, 知识生产, 知识灵韵, 生成式人工智能, 平庸之恶

Abstract: Large language models are intervening in knowledge production on an unprecedented scale. Their impact extends beyond mere efficiency gains to fundamentally reshaping how knowledge is understood, evaluated, and transmitted. Their generative mechanism, centered on intertextuality, enables texts to circulate and reconfigure frequently within existing discourses, yielding highly fluid, logically coherent, and widely acceptable forms of knowledge. Yet within this seemingly “rational” generative process, differences, negations, and critical judgments are progressively compressed. Complex issues are prematurely converged into safe, stable answers, revealing a latent and structural tendency toward the “banality of evil” in knowledge production. This banality is not about errors or low quality knowledge, but rather the automation, scaling, and subjectlessness inherent in intertextuality—which shifts the burden of judgment and displaces the subject’s position. Consequently, knowledge production increasingly relies on algorithms probabilistically fitting existing corpora, no longer unfolding through debate, hesitation, and reflection. Instead, it can be continually confirmed and reinforced along predetermined paths of technological intertextuality. If this persists, the unique spiritual resonance, humanity, emotion, and experience inherent in knowledge production will gradually be extinguished. Therefore, the true challenge brought by technological progress lies in preventing knowledge from losing its tension and depth during continuous generation. Only by reserving space for uncertainty and multiple interpretations in technological design, preventing efficiency logic from imposing low-level lock-in on innovation pathways through institutional arrangements, and reemphasizing the value of judgment, responsibility, and intrinsic insight at the cultural level can knowledge production escape the gravitational pull of mediocrity. It can then reclaim its role as a process leading to new questions, new meanings, and new possibilities.

Key words: intertextuality, knowledge production, knowledge’s spiritual resonance, generative artificial intelligence, the banality of evil