面向大语言模型的知识实践  

K nowledge Practices Facing Large Language Models

作　　者：吴冠军[1] Wu Guanjun

机构地区：[1]华东师范大学政治与国际关系学院

出　　处：《学术前沿》2023年第21期45-56,共12页Frontiers

基　　金：国家社会科学基金重大项目“后现代主义哲学发展路径与新进展研究”的阶段性研究成果,项目编号:18ZDA017

摘　　要：在同大语言模型知识实践的并置中,我们可以定位到学科交叉融合的必要性。以ChatGPT为代表的大语言模型,尽管才刚刚进入人类的视野中,但已经在知识实践上展现出卓越能力,成为堪称“通”家的大“专”家。我们可以用“模拟模式”与“数字模式”来分别描述人类与大语言模型的知识实践。大语言模型问世前的人工神经网络算法,数字模式的知识实践仅仅令其在狭窄的垂直领域展露出卓越智能。然而以海量人类文本为训练数据的大语言模型,其知识实践则呈现出无视领域疆界的通用性。面对大语言模型在知识实践中的应用,人类何为?潜在论与量子物理学,给出了我们积极展开跨学科知识实践的理据。In the juxtaposition with large language models'knowledge practices,we can locate the necessity of interdisciplinary integration.Large language models,represented by ChatGPT,although having just entered our horizon,have already demonstrated exceptional capabilities in knowledge practices,becoming akin to"generalists"in expertise.We can characterize the knowledge practices of humans and large language models respectively as"simulation mode"and"digital mode",Before the advent of large language models,the digital mode of knowledge practices in artificial neural network algorithms only allowed them to exhibit superior intelligence within various narrow vertical fields.However,large language models,trained on massive amount of human texts,exhibit a generality that disregards domain boundaries.Faced with the overwhelming superiority of large language models in knowledge practices,what should humans do?Potentiology and quantum physics provide the rationale for actively engaging in interdisciplinaryknowledge practices.

关 键 词：大语言模型 后人类 模拟模式 数字模式 纯粹潜能 量子思维 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] C19[自动化与计算机技术—控制科学与工程]