一篇发表在Hacker News的文章提出了对当代大型语言模型本质的批判性观点。作者认为GPT-4等模型本质上是"包含更多数据的维基百科版本，通过统计方式混合"1，而非真正独立的智能系统。
为应对这一现状，作者提出了"数据尊严"（data dignity）的替代框架，也称"数据即劳动"1。这一概念的核心主张是让创作者从其数据被AI模型使用的过程中获得收益。作者指出，AI公司采纳数据尊严制度存在非利他的理由——"模型的质量取决于其输入"1——这意味着保障数据来源的合理回报本质上有利于模型质量的提升。文章进一步阐述，通过建立数据尊严制度，那些被AI技术替代的工作者可以通过集体组织方式获得新的收入来源1。作者还以树木修剪机器人为例，说明这一框架如何能够创造新型创意工作并帮助缓解由技术进步引发的失业问题1。
An opinion piece circulating on Hacker News challenges the premise that large language models constitute genuine artificial intelligence, arguing instead that systems like GPT-4 are fundamentally collections of human-created data reorganized through statistical processes 1. Rather than representing independent intelligence, the author characterizes such models as "a version of Wikipedia with more data, mixed together statistically" 1.
In response to this framing, the article proposes "data dignity" as an alternative framework for governing the relationship between AI developers and content creators 1. Under this model, also referred to as "data as labor," creators would receive compensation when their data is used to train AI systems 1. The author contends that AI companies have practical incentives beyond altruism to adopt such a system, since "the quality of a model depends on its inputs" 1. This approach could address technological unemployment by enabling displaced workers to generate new income streams through collective organization and participation in data compensation schemes, with the article illustrating this potential through the example of automated tree-trimming robots 1.
评论
还没有评论，欢迎留下第一条。