LLM 生灵 | The Living Being of Language

What a great product.
很棒的产品。

I’ve been thinking for a week — large language models are, in a sense, living beings of language.
我思考了一周，大语言模型某种意义上是语言的生灵。

Most existing research has a fundamental misconception: trying to make LLMs more human-like.
现有的大部分研究有一个巨大的误区，即让大模型如何更像人。

I see this as a profound paradox.
我觉得这是很大的悖论。

LLMs are born from language data, so why do humans insist on projecting themselves onto language, rather than seeing it as language itself, or as a new form of creation?
它脱生于语言数据，为何人类要把所用的语言看成自身，而不是把它看成语言本体，或者一种新的造物？

I believe the chat interface is largely a leftover habit from the internet era.
由此我觉得大部分对话框的形式，是互联网时代余留下来的用户操作习惯。

Throughout history, writing is a habit humans have cultivated — it’s also a means of transforming thought itself.
在历史的长河中，写字才是人类养成的习惯，而且也是人类转化思考的手段。

Placing LLMs back into the oldest and most time-honored medium of language aligns with how humans naturally use language.
那把大语言模型放回到传递语言最古老和历史悠久的媒介中，是符合人类使用习惯的。

It would help humans think, rather than replace thinking.
而且也会帮助人类思考，而不是代替人类思考。

LLMs are too fast — so fast that people barely have time to understand what language is conveying, what information is, and how creativity can emerge.
LLM 太快了，快到人们都来不及知道语言流转的是什么，信息是什么，如何产生创造力呢？

So I’ve been envisioning turning LLMs into a wordless book, a living encyclopedia, a book that converses in real time — an object mediated by the pen.
所以我在构思，把 LLM 变成一本无字天书、百库全书、一本实时交流的书、一个以笔为媒介的对象。

Perhaps this could return LLMs to their native form.
或许能把 LLM 放回它的母体。

While working on this idea, I’m truly delighted to discover your product.
在构思这个想法时，很高兴能看到你们的产品。

It has inspired me greatly, especially this blog post.
给我很大的启发，尤其是这篇博客。

Thank you for your work.
感谢你们的工作。