Open source agency KAIYUANSHE
Recently, the WeChat public account subscription function has been adjusted!
Accounts that have not been starred may not display the big picture in the news feed!
Quickly star ⭐ us, and you can see the published articles in time!
STEP01 Click on the logo in the upper right corner
STEP02 Click 【Set as Star】
origin
In 2015, I wrote an article "How to Evaluate a New Technology——Taking Docker as an Example". Today, I plan to evaluate the recently popular ChatGPT in a similar style.
Eight years ago, I made a very high evaluation of Docker, and I think Docker is a more important technological advancement than Java. Now it seems that there is basically nothing wrong. Docker opened the era of cloud native, prompting Google to open source K8S. Although the company Docekr seems to be no longer important, container technology, cloud native technology, has changed the entire industry.
Now, ChatGPT has appeared, and at a faster rate than before, it is popular, popular, and out of the circle, which has aroused capital pursuit and various worries. It seems that the whole world is discussing it. Of course, there are also some disapproving remarks, thinking: it cannot be a myth, it should not be overestimated, it is nothing more than that and so on. So, how should such a new technology be viewed?
First of all, what exactly is ChatGPT?
It was relatively simple to evaluate Docker before, because at that time everyone could understand at a glance what Docker was. But what exactly is ChatGPT? I have recently intensively read many articles, many introduction videos, and participated in many rounds of discussions on WeChat groups and Twitter. The biggest feeling is: "The blind man feels the elephant".
There are various metaphors, and the more famous ones include: general artificial intelligence (Zhang Junlin), which has already begun to take shape, fuzzy lossy compression (Ted Jiang), plagiarism expert (Chomsky), and the browser just born in the Internet age. Machine (Wang Jianshuo), Guessing expert (WeChat public account of N articles) and so on.
After reading so many articles about blind men and elephants , although I still don't know how to define/locate ChatGPT, at least I know: ChatGPT is at the level of elephants.
What characteristics of ChatGPT do I pay attention to
There are many screenshots of conversations with ChatGPT circulating on the Internet, which everyone loves to hear. After all, whether it is amazing or funny, boring or superficial, it is a new thing worth watching. However, we should ponder a little deeper and look at the appearance of those dialogues. What important features do they embody? And these features are not available in previous conversational AI.
01
dialogue ability
On February 12th, I was discussing with people on twitter, and I had a guess: the unit of ChatGPT learning should not be "text", but "multiple rounds of dialogue in one scene", so his natural dialogue ability is already very strong , after more than a month of human training, he will definitely become stronger.
However, some of the views behind this cannot be expanded on twitter.
I once proposed a source of machine learning corpus, which is PR in the open source community, including: code submission Diff, continuous improvement around various comments, and then whether the code is incorporated. Such a discussion thread has code, natural language, comments on specific lines of code, changes before and after the code, and final judgment
result. If the machine can learn such material, it will naturally be able to provide better programming suggestions.
Promote various materials on social networks. In fact, BBS, Blogs, mailing lists, various chat rooms, social networks, etc., are having conversations everywhere.
These dialogues are not of a single content, single scene, and single language (it is likely to be mixed, and even constantly digress)
These dialogues show the process of human beings understanding each other, expressing themselves, and communicating with each other
After the launch of ChatGPT, this process took place between man and machine. Every conversation (this is very clearly divided on the ChatGPT interface), and every human comment on the conversation (like button), is people helping ChatGPT continue training
People correct the content of ChatGPT in the conversation, which is also the training input
What we've been talking about is AIGC, talking about how ChatGPT generates content. However, what should really be concerned is: ChatGPT's dialogue ability is enhanced using AI.
02
The Power of Multilingual Models
I once asked ChatGPT a question: Can the increase in English corpus help improve your Chinese ability?
His answer is: Yes, the increase of English corpus can also help ChatGPT improve its Chinese ability. As a multilingual language model, ChatGPT can learn the grammar and semantics of multiple languages at the same time. If ChatGPT is exposed to a large amount of English corpus during training, it will learn English grammar and semantics, thereby enhancing its ability to understand the Chinese language. Note that the multilingual ability of the model is different from its language translation ability, and it still needs to enhance the translation ability through a large amount of professional translation corpus.
After reading such answers, we can pay attention to and analyze the following contents:
ChatGPT uses a multilingual model
In fact, it is not just a variety of natural languages, but a variety of natural languages, plus a variety of programming languages, plus a variety of "artificial languages" (such as: Markdown, XML, CSS, CSV, etc.)
So many kinds of languages, within GPT 3.5, should be able to promote and reinforce each other as a whole
The mixed input and output of code and language, and the translation between multiple languages are all the capabilities of ChatGPT based on a unified model
For example: the translation of ChatGPT is not "accurate translation", but "free translation", which is actually more powerful
Pay attention to the last sentence of his answer: "The multilingual ability of the model is different from its language translation ability. It still needs to enhance the translation ability through a large number of professional translation corpora."
Among the various language materials used by ChatGPT, it is classified
Different types of corpus have their own specialties, but they can reinforce each other
This sentence is what he thought of to say. My question does not actually include the question of translation ability. I can only guess that in other rounds of conversations, countless people have asked where ChatGPT's translation ability comes from.
03
general intelligence
Dialogue between humans can actually discuss anything. However, in past human-machine dialogues, we tend to presuppose that this AI can only answer certain kinds of questions. However, since the corpus of ChatGPT may be any existing instance of human dialogue, we can also expect that various types of dialogue can be tried to communicate with ChatGPT.
As for the effect of human-computer communication, it depends on the quantity and quality of the corpus of this type of dialogue. So, now everyone is exploring what to chat with ChatGPT. It's going to be fun, and possibly "scary".
I agree with Zhang Junlin's judgment: ChatGPT can already be considered a kind of general artificial intelligence. It is because of this general dialogue ability, in short: he can chat with you about everything, and there is a high probability: these chats are still worthwhile. No matter how we evaluate his "routines", at least he is not talking "empty words", not like some chatbots, who only talk or talk nonsense.
04
Fill in the blanks and inspire thinking
A few days ago, I chatted with ChatGPT about open source studies. To be honest, I was very surprised. Because: there is no open source science at all, at most we are a group of open source people, in a very small circle, doing some early attempts.
In my opinion, this means that ChatGPT is able to fill in the gaps. Between the disciplines A and B that humans have developed, there is the possibility of cross-disciplinarity, and such a possibility, through communication with ChatGPT, can inspire thinking, which is very valuable.
Three elements to evaluate a new technology
In the article 8 years ago, I actually tailored three standards for Docker, and now I can also try to see if I can apply it to evaluate ChatGPT.
01
Improve efficiency
Of course, ChatGPT has greatly improved our work efficiency, and it is very obvious when many of us had to do boring and trivial things by ourselves before (for example, writing weekly newspapers). Of course, there are still people who use him to do various jobs and have achieved more or less results.
However, ChatGPT will not be responsible for the authenticity of the content, so if you don’t check it yourself, or even deliberately fake it, it will also bring complicated and difficult authenticity problems.
02
increase choice
Before ChatGPT appeared, some of the things we were doing, such as searching, learning English, and chatting with friends, may now only need to chat with ChatGPT. Of course, after the launch of Bing+ChatGPT, such a search may turn people from Google to the embrace of Bing.
Indeed, there are more choices than ever (not good news for Google)
03
lower the threshold
ChatGPT's ease of use and wide range of uses have lowered the threshold for using AI. Although the fastest to reach 100 million users, it may be a false (over-exaggerated) number. However, it is an indisputable fact that overwhelming users flood into ChatGPT and start chatting with it.
However, such an evaluation standard is simply not enough to measure its importance for ChatGPT.
Another set of three elements for evaluating a new technology
This is actually the last digression in the original article. It is a bit too much to evaluate Docker, but it is just right for ChatGPT.
01
From Quantitative Change to Qualitative Change
Docker has actually caused a qualitative change, and the entire IT R&D and DevOps have changed accordingly. It is conceivable that ChatGPT will change the nature of artificial intelligence and even the nature of the IT industry.
02
Create a new industry, and even more industries
There are not many industries created by Docker, maybe YAML configuration engineers are one. However, we can foresee that the API provided by ChatGPT can lead to more AI applications, and this field will be unimaginably vast.
03
Hazardous
The harm of AIGC has entered a new stage after the emergence of ChatGPT that can be faked
A number of scientific journals have explicitly prohibited the writing and submission of papers involving ChatGPT
There are already experts discussing very seriously: AI unemployment
Summarize
The emergence of ChatGPT is a far more important event than the emergence of Docker. Perhaps a signal of the "new industrial revolution" level! The next node, perhaps, is that AI can find a way: self-training, self-tuning, self-evolving.
Since GPT 4 and even GPT 5 are not the ceiling for the development of AI technology, in the next 5 to 10 years, we need to pay very, very attention to this field, and the importance of all other technical fields must be placed behind.
PS. Technologies such as blockchain, metaverse, and Web3 are really not that important.
Author | Zhuang Biaowei
Editor | Li Jiayang
Related Reading| Related Reading
Open source enterprises go overseas｜2023 "lion" must go, take you to see Singapore's science and technology ecology
Hackers and Customers: Can Open Source Software be Commercialized?
Introduction to Kaiyuanshe
Founded in 2014, the Open Source Club is composed of individual members who voluntarily contribute to the open source cause. It is formed according to the principles of "contribution, consensus, and co-governance". It has always maintained the characteristics of vendor neutrality, public welfare, and non-profit. International integration, community development, and open source projects" is an open source community federation with the mission. Kaiyuanshe actively cooperates closely with communities, enterprises and government-related units that support open source. With the vision of "Based in China and Contributing to the World", it aims to create a healthy and sustainable open source ecosystem and promote China's open source community to become an active force in the global open source system. Participation and Contributors.
In 2017, Kaiyuanshe was transformed into an organization composed entirely of individual members, operating with reference to the governance model of top international open source foundations such as ASF. In the past nine years, it has connected tens of thousands of open source people, gathered thousands of community members and volunteers, hundreds of lecturers at home and abroad, and cooperated with hundreds of sponsors, media, and community partners.
