大语言模型的真实与虚拟

The Reality and Virtuality of Large Language Models

本论文旨在厘清当今商业化的大语言模型带来的真实与虚拟混淆现象。首先,从技术内部视角出发,沿着人工智能发展史溯源,明晰德雷福斯与明斯基等人的分歧。其次,从数学尤其是计算科学中的随机性问题入手,探讨在早期人工智能流派逐渐式微后,辛顿等人的人工神经网络改良工作如何让行业重整旗鼓,同时又如何埋下了掩藏真相的隐患。再次,结合鲍德里亚对海湾战争的大众媒介批判和基特勒的软件批判,论述美国意识形态如何通过拟像构建共识,以及这种基于虚拟的权力结构在人工智能时代又是如何进一步加固自身的。而在拟像世界愈发根深蒂固、物质世界逐渐沦为幻象的今日,依然可以从媒介技术层面即编程语言史中找到一种突破虚拟交互界面的可能性,并发现从面向机器编程到面向过程与面向对象编程逻辑变化中暗含着的主体性变迁和用户层面的可操作性。最后,呼吁在人工智能的超现实未来到来之前,对当前的共识进行反思和重塑,以期使科技环境更加透明和真实。

This paper aims to clarify the confusion between reality and virtuality caused by the commer⁃cialization of contemporary large language models.From an internal perspective of technology,the paper firstly traces the historical development of AI,elucidating the divergences between scholars such as Hubert Dreyfus and Marvin Minsky.Then it examines issues of randomness in mathematics,especially in computational sci⁃ence,and explores how the improvement work on artificial neural networks by Geoffrey Hinton and other re⁃searchers revitalized the field after the decline of early AI paradigms,meanwhile also sowing the hidden dan⁃gers to conceal the truths.Further,the paper combines Jean Baudrillard′s critique of mass media during the Gulf War with Friedrich Kittler′s critique of software to discuss how American ideology constructs a consensus through simulation and how this power structure,based on virtuality,is further reinforced in the age of artifi⁃cial intelligence.In an era where the virtual world becomes increasingly entrenched and the material world fades into illusion,this paper argues that it is still possible to find a way to break through the virtual interface by examining the history of programming languages from a media technology perspective.This examination re⁃veals an underlying shift in subjectivity and user operability from machine-oriented to process-oriented and object-oriented programming logic.Finally,the paper calls for a critical reflection and reshaping of current consensus before the advent of a hyperreal future dominated by artificial intelligence,with the aim of making the technological environment more transparent and authentic.