definition
What is ChatGPT?
ChatGPT is an artificial intelligence chatbot program developed by OpenAI and launched by OpenAI in November 2022. The program uses a large language model based on the GPT-3.5 architecture and is trained by reinforcement learning.
ChatGPT interacts in a conversational manner and can be used for a variety of tasks including automatic text generation, automatic question answering, automatic summarization, etc. For example, in terms of automatic text generation, ChatGPT can automatically generate similar text based on the input text, and in terms of automatic question answering, ChatGPT can automatically generate answers based on the input questions.
This is ChatGPT's own introduction:
Note: Self-introduction is good, but "transfer learning" should be translated as "transfer learning" to be more accurate.
The technology behind ChatGPT is OpenAI's GPT large-scale language model framework, and OpenAI has also received significant investment from Microsoft. ChatGPT answers the question "How do you analyze Microsoft's continued investment in OpenAI?":
Microsoft's continuous investment in OpenAI may be due to the following reasons:
Advanced technology: OpenAI is an advanced artificial intelligence research institution, and its research results and technologies are highly innovative and potential in the field of artificial intelligence. Microsoft may hope to be able to invest in access to these advanced technologies.
Market Opportunities: As AI technologies continue to develop, so do AI market opportunities, which Microsoft may hope to capitalize on by investing in OpenAI.
Cooperation opportunities: OpenAI and Microsoft are highly complementary in the field of artificial intelligence, and Microsoft may hope to
Well, chatGPT didn't finish the last sentence, nor did it explain it.
There are various things that chatGPT can do on the Internet, and our bloggers have also written a lot of blogs: https://blog.csdn.net/nav/advanced-technology/chatgpt
In my opinion, there are several reasons why GPT broke the circle and set off a popular upsurge:
- Able to handle context like a human, and solve problems with users step by step.
- Being able to talk in natural language reduces the threshold of AI to 0. This breakthrough may not matter to IT professionals. In fact, this breakthrough has crossed the "chasm of innovation" (see the figure below). The various AI breakthroughs in the past have aroused great interest in the industry, but they have not really touched the public. This ChatGPT allows most users (Early Majority) to personally experience the benefits of AI for themselves. Help, so to speak,
星星之火已经出现燎原之势.
ChatGPT can generate a variety of content and can answer many technical questions. Everyone’s attitude towards it was curious and amazing at the beginning, and then there were different reactions, especially in the field of Q&A and education –
Restrict, ban chatGPT
StackOverflow, the largest question-and-answer platform in the English IT world, has banned the use of ChatGPT-generated content on its platform. The main reason is the answers generated by ChatGPT 看起来非常高质量，非常容易生成，但是准确度低. This leads to a lot of users mass producing such content, which is detrimental to both the platform and the users who want to get the right answer.
This is a screenshot of the explanation on StackOverflow's official website:
A user (Alex Poole)'s comment got a lot of likes: (I translated it into Chinese)
Regardless of whether the answer generated by AI is correct or not, "StackOverflow is a website for professional programmers and enthusiastic programming enthusiasts. Q&A site" This sentence applies to both the [Ask] and [Answer] sections. In my opinion, people who just paste their questions into the AI and then copy the AI's answers back are not professional programmers, nor are they enthusiastic programming enthusiasts. Maybe these people have checked and tested the AI's answer, and if they can really verify that it's a good correct answer, they should be able to write it themselves. That doesn't necessarily mean there isn't (or won't be) a place on the internet for AI, but it doesn't belong on StackOverflow.
Here is the translation I got using deepl.com service:
Some schools in the United States also prohibit students from using the school's network to connect to ChatGPT. The main reason is: "While the tool may provide quick and easy answers to questions, it does not develop the critical thinking and problem-solving skills that It is crucial to academic and lifelong success" ( report link ) However, the school cannot prevent students from connecting to ChatGPT in other ways, so that those who have mobile data can use it, and those who do not have mobile data cannot use it. Or, a small number of students in a certain area who can access the Internet in a certain way can get this kind of service. Does this create an unfairness?
What if something goes wrong?
About 10 years ago, my colleagues in the research institute developed an AI writing assistant. You input the previous sentence, and the writing assistant will suggest the next sentence according to the context. At that time, the AI technology and data volume were still relatively rudimentary, so its Some suggestions seem childish and ridiculous. After a small-scale demonstration of this research project, it stopped. Ten years later, it is very different. This is the topic I just randomly posed to ChatBot:
What if something goes wrong with ChatBot Chat? This is what actually happened. I tried to chat with her about the "testing on the edge of crime" joke, and ChatGPT was very calm:
In 2016, Microsoft released an emotional chatbot - Tay , but 16 hours later, it took her offline. Because in the process of chatting with users, she learned a lot of swearing, discrimination, and other bad behaviors from users. Does this chatbot have "consciousness" and "conscience", and does it know that its words will hurt others? Take the opportunity to ask her directly.
Judging from my limited tests, ChatGPT (the expert behind him) wisely learned the lessons of his predecessors, drew some red lines for himself, and dared not cross the threshold half a step.
coexist
ChatGPT can quickly give a seemingly confident answer, and students can just copy it. Is this cheating? Some teachers also said that this is like when Google appeared 20 years ago. At that time, some teachers did not allow students to find answers directly from Google. What happened later?
If you can't beat you, just join you.
This actually reflects artificial intelligence (simple or complex algorithms) 辅助人类工作的一个趋势. Here are some examples:
- Word editor's automatic spelling correction function (is this cheating)? And the continuous improvement of grammar suggestion function, Excel's automatic data filling and calculation function, PowerPoint's "slide automatic typesetting" function
- In the past, beginners had to look up a cardboard dictionary word by word when translating articles. Later, there were dictionaries on the Internet, portable electronic dictionaries, and mobile phone apps. Now, many translation tools can translate entire sentences with good results. Like the deepl.com example used above in this article.
- IntelliSence in the code editor automatically completes variable and function names, automatically generates comments for function APIs, evolves to the latest auto-generated test cases, automatically completes function functions, and more.
- Wolframalpha.com's automatic problem solving and listed steps to draw beautiful graphs…
- Beauty in artistic creation, AI painting, etc.
- Not to mention the role of AI algorithms in stock market trading...
These advances aroused some controversy at the time, but now they seem to be quite natural.
Of course, AI has developed from solving a single discrete task to being able to complete a set of exams, which is a huge leap - recently, the world-renowned Wharton professor Christian Terwiesch asked ChatGPT to do the final exam paper of the course , and the AI actually scored about A score of B or B-. Considering the speed at which AI is developing, it's not hard to imagine a not-too-distant future where AI crushes humans in MBA courses, just like AlphaGO crushed human Go players... So,上 MBA 学院的价值是增加了还是减少了呢?
In the interviews of our software engineers, we often have to examine basic algorithms, such as quick sort, red-black tree , etc. If a candidate says that these algorithms have APIs and other tested algorithm libraries, I just call them up! As an interviewer, what do you do? In exams and interviews, we often have to examine this person's "skills". So, is it a valuable skill to use ChatGPT? What is the opposite of skill ?
Encourage the use of ChatGPT
If you can't beat you, just join you, and make full use of this tool. I saw this report on npr.org today . A teacher asked all students to use ChatGPT for homework. He thinks that this tool is still very good in the discussion and creative stage.
The teacher said that he was excited and anxious about the impact of AI tools on teaching, but he still had to keep up with the times!
The famous mathematician Tao Terence imagined that the exam should move towards the "open books, open AI" mode - open the book and allow AI tools. In this case, how should our test paper be designed to distinguish the 好/中/差students ?
Clearly distinguish [reference] or [cheating]
When some people complain that ChatGPT leads to cheating, they may forget that cheating is a behavior that has existed in human history, and it is no exception now. You can search for "thesis writing " or " graduation project source code " - search for these things People, do you collect data, or take ready-made plans as your own works? This is unknown.
This BBC report also mentions large assignment writing services in Kenya. Once the cheating started, everyone stepped down from the high ground of "academic morality", followed by cheating in various links:
- These homework writers are all African compatriots in Kenya, but they all have white faces
- The five-star praise on these homework assignment accounts is also fake, and there is also a trading market for fake accounts and fake reviews
- This gave customers (students in Europe and the United States) an impression that "high-level Westerners are writing articles for me, and my money is worth it"!
From the data in the report, it can be deduced that a ghostwriter who works seven days a week can earn an average wage equivalent to that of a local adult. This is a very good way out for many young people who have received higher education but cannot find jobs in the local area.
Of course, people have also developed various tools to detect cheating. For ChatGPT, there is also a detection tool called GPTZero !
If everyone can use ChatGPT, and the technology to detect ChatGPT is also improving, then, what impact will it have on the human flesh essay writing industry in Kenya? Can we train ChatGPT to detect cheating?
AI needs self-knowledge
How does ChatGPT see that the content it generates is used by students? See its direct answer:
In theory, which level of AI does ChatGPT belong to? If the low-level AI is so powerful and the advanced AI appears, what should we do?
(Picture from: Professor Wang Jun's speech )
Of course, the competition between humans and AI in various fields has just begun. The magic height is one foot, and the road is one foot high. The process and result of the contest between the two sides will be cooperation, coexistence, integration, or confrontation? Who will win in the end? Let's 拭目以待(speaking of waiting and seeing, maybe someone is experimenting with machine eyes plugged into the human brain)!
Follow-up... Greater Fallout
Recently, some authors listed ChatGPT as a collaborator of their articles and published academic papers, which aroused the concerns and debates of many experts. After writing the blog post, I also saw such reports on the Internet: The Guardian ’s Reports and explanations from other netizens: https://weibo.com/1924545467/MqwNKgeKw
The core decision is:
主要科学期刊出版商决定禁止将 ChatGPT 列为署名作者，但是允许真人的作者在论文准备阶段使用 ChatGPT 等工具，只要说明使用的情况即可。
ChatGPT has also caused ongoing debate in the global education community. According to the information collected in this article , we can see:
About 英语教育Herman, a San Francisco high school English teacher with 12 years of teaching experience, believes that ChatGPT can complete the writing of the main text types in an instant, which subverts the basic functions of English teaching, and the status of English education is also in jeopardy. Pondicio, a senior researcher at the American Enterprise Institute, believes that Herman's views only represent part of the intellectual elite, and all kinds of thinking training advocated by the intellectual elite must be based on a considerable knowledge reserve. Since the knowledge and language ability of most students have not yet reached the basic level, writing training is still essential. Pondicio also satirized the current over-admiration of ChatGPT: "Rather than creating content, ChatGPT is better at filling the gaps in the conversation through fabrication and rhetoric; this software is really good at imitating human nonsense."
About 技术与人文之辩Canadian writer March believes that compared with the rapid advancement of science and technology, humanities scholars have long been obsessed with the construction and criticism of grand narratives, but neglected practical issues such as language, history, ethics, and politics. While new technologies such as ChatGPT are becoming more and more good at completing "liberal arts work" such as dissertations, the gap between humanities and technology will continue to widen. However, Ream, a Forbes columnist, believes that the impact of ChatGPT on papers and even humanities education cannot be overly exaggerated-in terms of paper writing, the most important thing is students 不断思考和尝试创作的过程. Humanities scholars should learn to navigate new technologies and teach students to use these tools skillfully.
Teacher Yu Shiqi from Southern University of Science and Technology also shared her feelings on Weibo :
I'm teaching "C/C++ Programming" this semester. I just assigned my first homework, and then I input the homework requirements into ChatGPT. After reading the results, I feel very mixed. In the first step, the program compiled by ChatGPT has achieved the basic requirements. Although the advanced requirements have not been realized, implementation suggestions are given. So I let it optimize and it gave a good solution. In the third step, at my request, it wrote a pretty good report. In my opinion, how to evaluate students faces a huge challenge!
Human beings have been learning mathematics for a long time. If you use the scene of selling groceries to calculate money, you have roughly experienced it. 在地上写写画画 - 石板上写写画画 - 算筹 - 部分人的心算 - 珠算 - 计算器 - 集成到其他工具中Now when you go shopping, you just keep scanning the QR code with the goods, and then you also scan your own wx/Alipay to pay. Figure out how to get change. But should teenagers learn arithmetic? want! Why? Because mathematics is a tool for understanding the world and exercising the analytical ability of the brain. Without this ability, human beings cannot perform other higher-level 思考、尝试、创作processes Arithmetic begins.
The same goes for programming languages and software engineering. ChatGPT provides us with convenient tools, allowing us to see many problem-solving methods and many preliminary results faster, but our own brains still have to go through a long learning process to realize it in the field of programming and software engineering. 思考、尝试、创作的过程, The challenge faced by teachers of IT majors is how to continue to create such a training environment in the ever-changing tool advancement, so that students can learn 思考、尝试、创作.
Reference article:
The Heart of Teaching: Building a Natural, Critical, Learning Environment