Wittgenstein and LLMs

查看原文

模型发布

热度趋势

新上榜

最近 24 小时与此前 24 小时对比 · 7 天曲线

百分比基于当前可用热度信号，而非评论数或独立用户人数。

AI 摘要

一位Reddit用户发帖询问维特根斯坦在《哲学研究》中提出的“语言游戏”概念与大型语言模型（LLMs）发展之间的关联。该用户提到，有说法称维特根斯坦的理论——即语言的意义基于“游戏规则”而非抽象含义，强调语言的语境——为LLMs的开发和训练提供了启发。用户希望了解这种影响是刻意的设计还是训练过程的自然结果。

Sorry if this is the wrong sub. I may crosspost to Ask Philosophy. I just have heard that Wittgenstein's concept of language-games - the meaning of language is based on "rules of the game," so the context of language, not abstract meaning - from Philosophical Investigations has informed the development and training of LLMs. I'm not sure if this was intentional or simply an outcome of the training.

Is anyone here familiar with this link or involved with the training of LLM that uses this concept?