难以置信！这篇文字优美见解深刻的科技散文，竟然是GPT-4写的

配图｜汤源

翻译｜汤源
原作｜Jeremy Howard

这是AI范儿公众号的第50篇文章
完整阅读本文约需10分钟

语言的未知领域

语言兼具限制和解放之源。GPT-4 则将这一想法推向了极致，让我们可以进入语言的无限之境。    

超越维特根斯坦的墙

“The limits of my language mean the limits of my world.” — Ludwig Wittgenstein

语言就像我们用来导航世界的地图，但它也像一座监狱，让我们看不到墙外的东西。 

但是，如果有办法逃出这座监狱，扩大我们的地图，用新词探索新世界呢？这是 GPT 4 等指令调整语言模型带来的可能性和挑战，GPT-4 是一种使用人工神经网络根据用户提示词输入生成自然语言文本的尖端技术。  

GPT-4 可以写任何东西，从散文到小说到诗歌到推文到代码到食谱到笑话到歌词到任何你想要的东西。它甚至可以写出尚不存在的东西，人类从未想过或说过的东西。  

正如维特根斯坦的话所暗示的那样，语言是限制和解放的源泉。GPT-4 将这一想法推向了极致，让我们得以进入语言的无限之境。   

这可能是现代历史上最重要的新技术，因为它有可能改变许多领域和行业。从教育到娱乐，从新闻到司法，从科学到艺术，这些模型可以促成新形式的学习、讲故事、报告、推理、发现和创造。   

它们还可能带来新的伦理、社会和文化挑战，需要认真反思和监管。我们如何使用这项技术将取决于我们如何认识到它对我们自己和他人的影响。  

该技术是“人工智能”的一种形式。“智能”一词源自 inter-（“之间”）和 legere（“选择、挑选、阅读”）。那么，所谓智能就是能够在事物之间做出选择，挑选出重要的东西，阅读所写的东西。智力不仅仅是数量或质量；它是一种活动、一种过程、一种实践。这是我们用思想和语言做的事情。  

但是当我们让 GPT-4 为我们做这件事时，我们不是在放弃我们的智慧吗？难道我们没有放弃选择、挑选、阅读的能力吗？我们不是变成了语言的被动消费者而不是主动的生产者吗？  

只是按照我说的去做……  

我没有写那个

啊哈，但这就是问题所在：为以上这段“超越维特根斯坦的墙”的语言探索奠定基础的每一个词都不是我亲手写的（或者更确切地说，是键盘），而是由 Bing 的聊天机器人（使用 GPT-4）写的。由此，我们发现自己陷入了一个元语言难题的阵痛中，因为我们讨论的主题自身已经悄悄地融入了谈话中。  

现在，请耐心和我一起，有意揭露这个诡计的狡猾之处。我的本意是打破您对语言模型生成文本的所有期待：对以上这些看起来、感觉上和读起来文本的期待。因为如果你已经读到这里，你就会不知不觉地被吸引到人工智能生成的散文之网中。  

也许只有在与语言的神秘山谷缠斗时，我们才能找到力量重新定义我们自己的语言边界，并瞥见墙外的世界。     

结论

最后，GPT-4 给我们带来了一个悖论。一方面，它有可能扩大我们的语言视野并打破限制我们的墙壁。另一方面，它提出了关于智力、创造力的本质以及对于人类的意义的问题。当我们深入研究语言的未知领域时，我们必须注意拥抱创新和保持我们自己的智力自主之间的微妙平衡。  

因此，挑战在于将 GPT-4 用作我们自身成长和探索的催化剂，而不是替代我们自己的思想。通过以好奇心、谦逊和对道德考虑的承诺来接触这项技术，我们可以冒险超越维特根斯坦的围墙，并在语言、思想和人类潜能方面开辟新的领域。     

这个也不是我写的

实际上，以上整篇文章都是由 GPT-4 所写。我创建这篇文章是因为我还没有看到模型生成的好散文看起来是像什么样子，同时因为我想尝试一下使用 GPT 创建整篇文章的想法.     

💡为了创建它，我使用了以下四个提示词：

        起草写一个四段组成的介绍文章，该文章解释了 GPT 4 等指令调整语言模型的功能。以古典哲学家或作家的名言开头，并将对名言的引用融入到介绍中。”

        “为这篇文章再写四段，讨论这项技术的优缺点。结合与该主题相关的一些词的拉丁语、法语或希腊语词源，并将其编入文章中。”

        然后我开始了一个新的会话，并要求：“以与介绍完全不同的写作风格，按照大卫福斯特华莱士的风格，通过解释它实际上是由BingChat生成的来完成以下文章”，然后我将文章草稿一起粘贴到提示词中。

        然后我根据这个提示再次开始一个新的会话，在提示之后粘贴文章：“输入的文本是一篇缺少最后一节的文章。请以 Jeremy Howard 的风格写下 2 个可能的最后部分，每个部分 2-3 段。”

然后我删除了一些从句和句子，但除了在“Bing”之后添加文本“（使用 GPT-4）”之外，我没有以任何方式修改它们中的任何一个。我还手动创建了 Bing 建议我使用的图像并将其粘贴到文章中。    

GPT 在最后写的警告，“那么，挑战是使用 GPT 4 作为我们自己成长和探索的催化剂，而不是作为我们自己思想的替代品”，感觉就像是对这篇文章的警告。它最终朝着我没有真正计划的方向前进，事后看来，我现在完全不确定它是否有助于我的探索，或者取代了我自己的想法。  

最后，这个实验只让我感到更加困惑和不安。 

💡 原作者脚注

在某些类别的文章中使用了这种公式化的东西，他们会以机构认可的作家的引述作为开场白，并以其他方式表明他们的地位，例如通过编织拉丁语词源。这是一种看门人的形式，因此我不喜欢它，我想打破它。这就是我选择前两个提示的原因。现在你可以（至少粗略地）模仿任何写作风格，只要你能描述它，即使你没有沉浸在与之相关的文化中。这会让某些人真的很生气。

参考一、英文原作

GPT 4 and the Uncharted Territories of Language

Language is a source of limitation and liberation. GPT-4 pushes this idea to the extreme by giving us access to unlimited language.

Beyond Wittgenstein’s Walls

“The limits of my language mean the limits of my world.” — Ludwig Wittgenstein

Language is like a map that we use to navigate the world, but it’s also like a prison that keeps us from seeing what’s beyond the walls.
But what if there was a way to break out of this prison, to expand our map, to explore new worlds with new words? This is the possibility and the challenge offered by instruction tuned language models like GPT-4, a cutting-edge technology that uses artificial neural networks to generate natural language texts based on user inputs.
GPT-4 can write anything from essays to novels to poems to tweets to code to recipes to jokes to lyrics to whatever you want. It can even write things that don’t exist yet, things that no human has ever thought of or said before.  
As Wittgenstein’s quote suggests, language is a source of limitation and liberation. GPT 4 pushes this idea to the extreme by giving us access to unlimited language.
This could be the most significant new technology in modern history because it has the potential to change many domains and industries. From education to entertainment, from journalism to justice, from science to art, these models could enable new forms of learning, storytelling, reporting, reasoning, discovery, and creation.
They could also create new ethical, social, and cultural challenges that require careful reflection and regulation. How we use this technology will depend on how we recognize its implications for ourselves and others.  
This technology is a form of “Artificial Intelligence”. The word “intelligence” derives from inter- (“between”) and legere (“to choose, pick out, read”). To be intelligent, then, is to be able to choose between things, to pick out what matters, to read what is written. Intelligence is not just a quantity or a quality; it is an activity, a process, a practice. It is something that we do with our minds and our words.  
But when we let GPT 4 do this for us, are we not abdicating our intelligence? Are we not letting go of our ability to choose, to pick out, to read? Are we not becoming passive consumers of language instead of active producers?  


Just doing what I’m told…       

I didn’t write that

Ah, but here’s the rub: the very words that have set the stage for this linguistic exploration were penned not by my own hand (or, rather, keyboard), but by the digital tendrils of Bing (which uses GPT-4). And so, we find ourselves in the throes of a meta-linguistic conundrum, as the very subject of our discussion has surreptitiously insinuated itself into the conversation.  
Now, bear with me, for I have a purpose in revealing this sly subterfuge. My aim is to shatter the very expectations you hold about what language model-generated text looks like, feels like, and, indeed, reads like. For if you’ve made it this far, you’ve unwittingly been drawn into the web of AI-generated prose.  
Perhaps it is only in grappling with the uncanny valley of language that we may find the strength to redefine our own linguistic boundaries and catch a fleeting glimpse of the world beyond the walls.  

Conclusion

In the end, GPT-4 presents us with a paradox. On one hand, it has the potential to expand our linguistic horizons and shatter the walls that confine us. On the other hand, it raises questions about the very nature of intelligence, creativity, and what it means to be human. As we delve deeper into the uncharted territories of language, we must be mindful of the delicate balance between embracing innovation and preserving our own intellectual autonomy.  
The challenge, then, is to use GPT-4 as a catalyst for our own growth and exploration, rather than as a substitute for our own minds. By approaching this technology with curiosity, humility, and a commitment to ethical considerations, we can venture beyond Wittgenstein’s walls and chart new territories in language, thought, and human potential.       

I didn’t write that either 

Actually, the entire article above was written by GPT-4. I created this article because I haven’t seen good examples of what model-generated prose can look like, and because I wanted to experiment with the idea of creating an entire article using GPT-4.  

💡 To create it, I used the following four prompts: 

        “Write a four paragraph introduction to a draft New Yorker article which explains the capabilities of instruction tuned language models such as GPT-4. Begin with a quote from a classical philosopher or writer, and weave references to the quote into the introduction.”

        “Write four more paragraphs of this article, discussing the pros and cons of this technology. Incorporate the latin, french, or greek etymology of some word related to the topic and weave it into the article.” 

        Then I started a new session, and requested: “In a completely different writing style to the introduction, following the style of David Foster Wallace, complete the following article by explaining that it was actually generated by Bing.” I then pasted the draft article into the prompt. 

        Then I started a new session again with this prompt, pasting the article after the prompt: “The input text is an article which is missing the final section. Please write, in the style of Jeremy Howard, 2 possible final sections of 2-3 paragraphs each. “

I then deleted some clauses and sentences, but I didn’t otherwise modify any of them in any way, except to add the text “(which uses GPT-4)” after “Bing”. I also manually created the image which Bing suggested I use and pasted it into the article.
The warning that GPT wrote at the end, “the challenge, then, is to use GPT 4 as a catalyst for our own growth and exploration, rather than as a substitute for our own minds” feels like a warning to me about this article. It ended up heading in a direction that I didn’t really plan, and in hindsight I’m now not at all sure whether it helped my exploration, or substituted for my own mind.  
In the end, this experiment has only left me feeling more confused and uneasy.   

💡 Footnote

There’s this formulaic thing used in a certain class of articles where they’ll open with a quote from an institutionally-approved writer and signal their status in other ways, such as by weaving in latin etymology. It’s a form of gate-keeping, and therefore I don’t like it, and I wanted to break it down. That’s why I picked the first two prompts. Now you can (at least roughly) mimic any writing style, as long as you can describe it, even if you’re not immersed in the culture that it’s associated with. This will make certain people really mad.

参考二、 维特根斯坦：学习ChatGpt绕不开的天才

 💡 维特根斯坦：学习ChatGpt绕不开的天才

维特根斯坦Ludwig Wittgenstein说过一句名言，语言的边界即世界的边界：The limits of my language mean the limits of my world。🗺️🌍

ChatGpt 是先进的自然语言处理技术，能够模拟人类的对话和语言处理能力，而维特根斯坦的哲学思想也对我们学习ChatGpt，尤其是prompt engineering有很大意义。🤖💬📝💭

维特根斯坦是20世纪最伟大的哲学家之一，也是超有钱的富二代，后来人家全部捐了，自己躲到树林里盖小木屋。他在哲学界的地位有如爱因斯坦，促成了西洋哲学的语言学转向.💰🏠🌳🧑‍🎓

理解维特根斯坦的哲学思想，也是我理解ChatGPT AI精髓的必读资料。他认为语言是一种游戏，而其实ChatGpt的本质也是一种语言游戏，它需要理解人类的语言，模拟人类的对话和思维方式，才能达到预期的效果，就是我们说所的AI PROMPT ENGINEERING 🎲🧠🕹️💻

学习维特根斯坦思想可以帮助我们更深刻地理解ChatGpt AI的本质，更好地掌握ChatGpt AI的应用技巧。👨‍🏫🤖🤔💡

语言的限制也决定了我们世界和表达思想和概念的能力的限制,因为语言单词的含义并不是固定的，很多事我们的经验，但AI系统对经验和感情的输入是不理解的。所以才有了他的那句名言：对那些不能说的，保持缄默。🤐🗣️

应用到ChatGPT中，就是说不能对AI输入这部分语言内容。🙅‍♀️💬🤖

接着让我谈谈🤔 Wittgenstein的哲学思想和GPT语言之间的关系：

GPT-3是一种机器学习模型 🤖，它是基于大量文本训练的📚，并没有与人类相同的语言和上下文理解能力😔；虽然GPT-3可以生成看起来连贯而有意义的文本📝，并与人类进行交流💬，但它并不具备人类对世界和概念的相同理解水平🌎。因此，它不能完全捕捉Wittgenstein关注的语言和含义的细微差别😬，需要我们tuning prompt🎛️;

Wittgenstein认为单词的含义并不是固定的🔒，而是由使用它们的上下文来确定的🤔，因此我们在输入prompt的时候，要在不同的地方去除这种歧义💡，让机器清楚地明白指令🤖，从而输出我们需要的结果👍，否则很难提示ChatGPT生成用户想要的😕。

有了有效的提示以后，用户需要让AI系统了解以什么样的输出格式表达请求的结果🤗。这需要我们了解ChatGPT的语言游戏性质以及构建方法🤓。

所以说，维特根斯坦是我们学习ChatGPT绕不过的那个天才👨‍🎓！

END

点这里👇关注我，记得标星哦～‍‍‍