几年前，通用人工智能是一个遥不可及的目标。但今天，ChatGPT给所有人带来了一场狂欢。它的颠覆性，它进化的速度，让人觉得震惊、不可思议——一个月之前，它还只能识别文字，现在，它已经进化到ChatGPT4，甚至10秒就可以生成一个网页。
本周六，我们邀请韦青老师，微软中国CTO做客混沌。他将利用工程师的底层视角，为大家揭开人工智能的冰山一角，探寻ChatGPT为何全球爆火。
一、ChatGPT怎么突然就火了？
1. 山中方七日，世上几千年
GPT4的出现一下让人工智能成了最热门的话题。但GPT3的模型，实际上两年前就培训出来了，它底层的技术结构、技术原理也存在很多年了。
2023年1月，微软CEO Satya和Sam Altman（OpenAI CEO）的合照在网上很火，他们跟大家介绍了双方的合作和技术的进步。但其实在2016年，双方已经开始探讨了。
什么意思？ChatGPT是机器时代的冰山一角。当我们看到某些现象出现的时候，某种意义来讲，这件事已经结束了。所以，我们现在要做的是什么？
2. “我在讲给ChatGPT怯魅”
韦青老师说，我想特别强调一点，我们误以为机器能凭空学到知识，不需要人的参与。其实机器是不理解概念的，机器理解的是概率分布。语言，每一个字的出现，都是有概率分布的。它的答案是基于概率的一种推理，不是概念的推理，这一点我们务必要明白，这是祛魅的一个必要条件。Sam Altman（OpenAI CEO）能够坚定信念，其实是因为他对语言的特征，有深刻的理解。数字化转型网www.szhzxw.cn
我们给ChatGPT披了一张画皮，呈现出一个智者的形象。想象一下，ChatGPT是一个狐仙。我问，它答，你会觉着它是一个神人。但把这画皮一揭开，原来你提出的问题，就是一段指令，你还会觉得很神奇吗？
二、你能做出一个ChatCPT吗？
3. 冲着ChatGPT创新，等于拎包入住而没有自己的地基
所有人都在盯着技术，想再做一个ChatGPT。韦青老师说，我们其实忽略了最重要的一点，能够把技术开发出来，先得有一个思想。ChatGPT，实际上它只是一种提取已经被沉淀、被记忆的知识的一种交互方式，底下什么？是基础模型。像谷歌的Bird，Facebook的Lama。
为什么GPT现在那么火？它背后是OpenAI的技术信仰，再加上微软的Satya（微软CEO）、Kevin Scott（微软CTO）这些人的技术信仰的结合。这些人都相信，靠大量的语料和海量的计算，能够产生出对于知识的沉淀和应用。
4. 我们有没有对于工具、对于数字化技术的技术信仰？
这才是我们应该问自己的。数字化转型网www.szhzxw.cn
认知达到信仰的层面，才能够选一条没有人走过的路。尽管今天还走不通，但你就认为这条路可以走，因为第一性原理告诉你，这样做是对的，只不过人类的技术还没有发展到那一步，还证明不出来。
另一个例子是SpaceX。它成功了，大家都觉得太牛了。但当马斯克后来接受媒体采访时，他哭了。他说当时我认为火箭可以回收，但NASA、所有大咖级人物、科学家都跟他讲，这条路是走不通的。
OpenAI是一样的，我们认为它在2023年1月忽然出名，怎么可能？它背后有太多的辛酸、坚持、疑惑、摇摆，只不过坚持下来了。
三、AI会不会代替掉人？
5. 为什么OpenAI会提出UBI（全民基本收入）？
20年前谷歌刚成立的时候，只是个搜索引擎。你会感到很无厘头，一个搜索引擎，为什么把不做恶作为标准？现在大家明白了吧，在一个“比真还真的时代”，给你的信息当然可以作恶。
AI也一样。当就算在AGI上，大家的关注点，也都各有侧重，微软明确指出要做负责任的AI，谷歌说的叫不作恶。OpenAI提出UBI，全民基本收入（UBI，是指“无条件”地为所有个人定期发放一笔现金收入）。
Sam Altman在采访中说，需要新思想回答的三个问题是：如何分配通用人工智能产生的利润？如何分享通用人工智能的访问权？如何分担通用人工智能的治理权？数字化转型网www.szhzxw.cn
大家想一想，为什么Sam Altman想到了这三个问题？
6. 人时常忽略何以为人
韦青老师说，你觉得机器像人对吧？其实我要说，是我们太不像人了，所以才认为机器像人。何以为人？尤其是当我们知道自己有那么多的误区、偏差之后，如何去弥补、防范思维误区和偏差，让我们做得像个人？这是远比AI会不会代替掉人，更核心、更本质的一句话。
越是纷纷扰扰，越要守住第一性原理。这个时代，门开了一道又一道，但外界其实一直这样，是有风险的，只是我们没有跳出舒适区。
7. “煤气灯下”、 知识“肥胖症”、知识“智子”
韦青老师说，知其雄，守其雌。怎么做到？有三个词可以供大家参考——“煤气灯下”、 知识“肥胖症”、知识“智子”。
煤气灯下Gas Lighting是2022年，韦伯字典的全球热词，是上一世纪40年代一部电影的名称。电影讲述了，一位先生想谋取他太太的财物，一直在给太太洗脑。电影用蒙太奇的手法，煤气灯摇摇晃晃，让人感觉到晕眩，感觉到不确定。它为什么变成全球热词了呢？因为现在机器生成的虚假消息和错误消息，就如同摇晃的煤气灯，让我们失去了对什么是真实的客观判断。数字化转型网www.szhzxw.cn
我们都知道，吃垃圾食品会带来肥胖症，那知识“肥胖症”又是什么？《三体》里的智子，是一种源头上的约束，让新的知识无法产生。那人工智能又会如何反过来影响人类知识？
翻译：
A few years ago, general artificial intelligence was a far-fetched goal. But today ChatGPT is giving everyone a bash. It’s amazing how disruptive it is, how fast it’s evolving — a month ago, it could only read text, now it’s evolved to ChatGPT4, and it can even generate a web page in 10 seconds.
This Saturday, we will invite Mr. Wei Qing, CTO of Microsoft China, to visit Chaos. He will use the perspective of an engineer at the bottom of the artificial intelligence iceberg to uncover why ChatGPT has become such a global phenomenon.数字化转型网www.szhzxw.cn
How did ChatGPT suddenly become so popular?
mountain China seven days, thousands of years in the world
The advent of GPT4 has made artificial intelligence the hottest topic of all. But the model of GPT3 was actually trained two years ago, and the underlying technical structure and principles have been around for many years.
In January 2023, a photo of Microsoft CEO Satya and OpenAI CEO Sam Altman went viral on the Internet. They talked about their cooperation and technological progress. But in 2016, the two sides are already talking.数字化转型网www.szhzxw.cn
What do you mean? ChatGPT is the tip of the iceberg in the machine age. When we see something happen, in a sense, it’s over. So, what are we going to do now?
“I’m talking to ChatGPT.”
‘I want to emphasize that we mistakenly believe that machines can learn knowledge without human involvement,’ Mr. Wei said. The machine doesn’t understand concepts, the machine understands probability distributions. Language, every word, has a probability distribution. Its answer is a kind of reasoning based on probability, not conceptual reasoning, which we must understand is a necessary condition for disenchantment. Sam Altman (OpenAI CEO) is able to believe because he has a deep understanding of the characteristics of language.
We give ChatGPT a painted skin and present him as a wise man. Imagine ChatGPT as a fox fairy. I ask, he answers, and you think he’s a man of God. But do you still think it’s amazing when you peel back the paint and the question you’re asking is just an instruction?
Can you make a ChatCPT?
The innovation of ChatGPT is tantamount to carrying the bag and not having its own foundation
All eyes are on technology to make another ChatGPT. Teacher Wei Qing said, we actually ignore the most important point, to develop technology, first have a thought. ChatGPT, it’s really just an interactive way of extracting information that’s already been deposited and memorized. What’s the bottom? It’s the basic model. Google’s Bird, Facebook’s Lama.数字化转型网www.szhzxw.cn
Why is GPT so popular right now? It’s a combination of the technical beliefs of OpenAI and the technical beliefs of people like Satya (Microsoft CEO) and Kevin Scott (Microsoft CTO) at Microsoft. These people all believe that knowledge accumulation and application can be produced by a large amount of corpus and massive calculation.
Do we have a technical belief in tools and digital technology?
That’s what we should be asking ourselves.
To reach the level of belief, cognition can choose a road that no one has walked. Even though it’s not possible today, you think it’s possible because first principles tell you that it’s the right thing to do, but human technology isn’t there yet to prove it.
Another example is SpaceX. It worked, and everyone thought it was amazing. But when Musk spoke to the media later, he cried. ‘I thought it was recoverable,’ he said. ‘But NASA, all the big people, all the scientists told him it was impossible.’
OpenAI is the same, we think it suddenly became famous in January 2023, how could that be? There is too much bitterness behind it, insist, doubt, swing, but stick to it.
Will AI replace people?
Why does OpenAI propose UBI (Universal Basic Income)?
When Google started 20 years ago, it was just a search engine. You will feel very ridiculous, a search engine, why do not do evil as a standard? Now you understand that in an era of “truth-than-truth”, of course, information given to you can do evil.
The same goes for AI. When it comes to AGI, everyone’s focus is different. Microsoft clearly points out to be responsible AI, while Google says don’t be evil. OpenAI proposes UBI, a universal basic income (UBI, meaning a regular cash payment to all individuals “without conditions”).
Sam Altman said in an interview that three questions that need to be answered by new ideas are: How to distribute the profits generated by general-purpose AI? How do I share access to general purpose AI? How to share the governance rights of general AI?
Think about it. Why did Sam Altman come up with these three questions?
People often lose sight of what makes them human
Teacher Wei Qing said, you think the machine is like a person, right? In fact, I would say that we are so un-human that we think machines are human. What makes a man? Especially when we know that we have so many errors, deviations, how to make up for, prevent thinking errors and deviations, let us do like a person? This is more fundamental and essential than whether AI will replace human beings.
The more confusion there is, the more we must uphold first principles. In this era, doors open one after another, but the outside world is always like this, it is risky, but we do not jump out of the comfort zone.
“Under the gas light”, knowledge “obesity”, knowledge “wise son”
Wei Qing teacher said, know its male, keep its female. How? Here are three words for your reference — gaslight, knowledge, obesity, and knowledge, wisdom.
Gas Lighting is the 2022 Webb Dictionary word of the world, the title of a 1940s film. The film tells the story of a man who wants to seek his wife’s property and has been brainwashing his wife. In a montage, the gas lights wobble, making one feel dizzy and uncertain. Why has it become a global buzzword? Because now machine-generated misinformation and misinformation, like a flickering gas lamp, deprive us of our objective sense of what is true.
We all know that eating junk food causes obesity, but what is “obesity”? The Wise son in The Three-Body Problem is a constraint on the source head that prevents new knowledge from being produced. And how will AI affect human knowledge in turn?
本文由数字化转型网（www.szhzxw.cn）转载而成，来源：混沌学园；编辑/翻译：数字化转型网宁檬树。
免责声明: 本网站(http://www.szhzxw.cn/)内容主要来自原创、合作媒体供稿和第三方投稿，凡在本网站出现的信息，均仅供参考。本网站将尽力确保所提供信息的准确性及可靠性，但不保证有关资料的准确性及可靠性，读者在使用前请进一步核实，并对任何自主决定的行为负责。本网站对有关资料所引致的错误、不确或遗漏，概不负任何法律责任。
本网站刊载的所有内容(包括但不仅限文字、图片、LOGO、音频、视频、软件、程序等) 版权归原作者所有。任何单位或个人认为本网站中的内容可能涉嫌侵犯其知识产权或存在不实内容时，请及时通知本站，予以删除。
