We are in an era of technological innovation. Artificial intelligence has and will continue to have a profound impact on our lives, work, and even the way we think.ChatGPT As a pioneer in this field, it not only represents the current level of technological development, but also indicates possible future development trends.
Hello everyone, I am Kevin, an AI explorer. Although ChatGPT has been released for a while, I recently found that when communicating with many friendsFriends of 80% have not used ChatGPT yet, or have a shallow understanding of ChatGPTTherefore, I specially compiled this article today for your reference and study, and also for myself to review.
The value of this article
This is a relatively comprehensive and easy-to-understand popular science article on ChatGPT, which will help everyone fully understand the working principle, application areas, future prospects, and possible challenges and impacts of ChatGPT.
Suitable reading audience
This article is suitable for all friends who are interested in artificial intelligence. You may be a technology fan, a company decision maker, a teacher, or a student or professional who is worried about future career development.
Article Outline
1) What is ChatGPT?
2) What can ChatGPT do?
3) Future prospects of ChatGPT
4) Train your own AI
5) Disadvantages and Problems of ChatGPT
6) What jobs might ChatGPT replace?
1. What is ChatGPT?
In principle, ChatGPT is not new. The basis behind it is the natural language model GPT released by its parent company OpenAI a few years ago. Friends who know something about artificial intelligence must know that OpenAI released GPT-1 in 2018, which for the first time enabled artificial intelligence to "understand text" and write text like humans. In the following years, OpenAI successively launched GPT-2 and GPT-3, using larger parameters to achieve more sophisticated learning. In layman's terms, it makes artificial intelligence more "human-like."
ChatGPT is a product that is packaged, fine-tuned, and optimized based on GPT-3.5. In the past, GPT-3 was just a bottom-level engine that had to be connected to a software before it could be used. ChatGPT, it can be understood that OpenAI itself has made such a software for GPT so that users can use it directly.
That's why it's called ChatGPT, which means "GPT that can chat". So, what is GPT? Its full name is Generative Pre-trained Transformer. Simply put, its principle is: first provide it with a huge corpus (usually directly grabbed from the Internet), let the model break up, mark, and learn these texts through hundreds of billions of parameters, and build a complex prediction model. Then, based on this prediction model, determine which word a word should follow in this context. In this way, words are strung together one by one to form a paragraph or an article.
This prediction model is actually very similar to the learning and processing model of our brain. Therefore, this is also an important reason why it can be more "human-like". We can understand it with a three-layer model: the "unsupervised" pre-training self-learning principle behind this model that does not require human participation, or the process of allowing the model to "grow freely" like the brain, is its power layer; GPT-3.5 completed by this principle is a structural layer, which is the main basis for ChatGPT to work;
ChatGPT, which is formed by packaging and optimizing GPT-3.5, is an interactive layer. Its essence is to add a more friendly user interface and interaction method, so that individual users can use it more easily.
Summary in plain words: ChatGPT is actually based on the GPT model developed by OpenAI. After optimization and adjustment, in order to make it easier for users to use, OpenAI specially added a user-friendly interface for it, allowing this AI to "understand" and write text like humans, achieving more humane interaction.
2. What can ChatGPT do?
After understanding the principles of ChatGPT, we will naturally be concerned about a question: What can ChatGPT do at present?
- 2.1 Answering questions
This is probably the simplest application of ChatGPT. You ask it questions in the chat interface, such as "How to write a good WeChat article", and it will explain it to you in fluent language, explaining the concept very clearly. As far as my experience is concerned, the effect is very good, concise and clear, and the expression ability is better than many people.
- 2.2 Writing Content
You can ask it to write a piece of content according to your requirements. The quality is quite good. Based on this, there can be many targeted applications. For example: (1) You can ask it to write an email to tell the customer your quotation, and express it in polite and sincere language. The work it writes is guaranteed to be accurate in wording and fluent in sentences, which is better than what ordinary people write; (2) You can give it a topic, some background and guidance, and ask it to write a planning proposal, or even ask it to help you think of ideas and provide ideas and inspiration in various directions. This can help you save a lot of time and energy; (3) You can even ask it to write a paper according to the teacher's requirements. In fact, many students abroad have begun to use ChatGPT to complete their homework, and university professors have begun to "compete wits and courage" with ChatGPT, such as asking students to write on the spot, asking students to explain the meaning of each paragraph, etc.
- 2.3 Summary and Refinement
ChatGPT also has a very strong ability to summarize and refine the input materials. For example, you can input an article and ask it to summarize the main idea of the article; or input a discussion and ask it to summarize the main points and arguments. Now there are also some products that use the GPT behind ChatGPT to achieve the following effect: you input a video or a podcast, and AI helps you listen to it. After listening, it summarizes the main points of the video or podcast in concise language and presents it to you.
- 2.4 Generate Code
Many programmer friends praise ChatGPT for its strong coding ability. If you ask a question, such as "how to achieve XXXX effect", ChatGPT will tell you several feasible solutions and provide the main functions and algorithms of these solutions; if you ask it a question, such as "a bug occurs in XXXX situation, why", ChatGPT will tell you several possible reasons and tell you how to deal with each possibility.
In plain language, ChatGPT can not only answer questions and write content like a human, but also summarize and refine the input information. It also has powerful programming capabilities and can provide code solutions and methods for dealing with bugs. In addition, its capabilities are far more than these, and there are many other functions waiting for us to explore and discover.
3. Future prospects of ChatGPT
So, what kind of changes will it bring to our lives?
- 3.1 Search
Search 1.0 era: Before the invention of the Internet, our information acquisition can be said to be in the Search 1.0 era. At that time, we could only go to the library to look up information, take notes with cards and notebooks, and record by photography or handwriting, which was tedious, troublesome, and costly. Therefore, "doing research" became a high-brow thing that only a few people were able to do.
Search 2.0 Era: Later, with the advent of the Internet and search engines, a large amount of information was digitized and put on the Internet. The way we get information has been greatly broadened. Just by entering keywords, we can see the literature published in various journals, the materials stored in the library, and see other people's analysis, opinions and insights. We have ushered in the Search 2.0 era of information explosion.
Search 3.0 era: With the assistance of AIGC tools such as ChatGPT, we may usher in the era of Search 3.0. What is the urgent problem that Search 2.0 needs to solve? There is too much information. How to effectively filter, refine and integrate this information has become a difficult problem. But AIGC can solve this problem. We ask it a question and give it a keyword. It quickly integrates the relevant information in the knowledge base, summarizes and refines the answer that best meets my needs, and presents it to us.
The whole process is smooth, natural and comfortable. It saves us the time of shuttling back and forth and sorting out large amounts of information. It is extremely convenient - this is Search 3.0.
- 3.2 Reading
As mentioned earlier, ChatGPT has the ability to summarize and refine, which may completely change the way we read and learn knowledge.
Imagine this: How will we learn a new concept in the future? You ask AI questions, and it explains the concept to you directly and in detail. If you don't understand something, you can continue to ask, and it will explain it to you further. So, do you still need to look for information, read textbooks, and attend classes? No. AI is enough.
How will we read a book in the future? Perhaps you can import it into artificial intelligence, and it will help you summarize and extract the most valuable content in the book. It may be the author's point of view, it may be the methodology and steps, it may be the case or argument... Whatever you want, you can give it instructions, let it help you "dehydrate" and present a perfect summary.
How will we watch videos and listen to podcasts in the future? Perhaps we don’t need to watch or listen anymore. We can just import artificial intelligence, which will help us summarize the key points, or even write a witty and literary reading report for us, so that we can easily and happily absorb knowledge.
- 3.3 Writing
This may be even simpler. We no longer need to carefully write emails ourselves, we only need to give it an instruction, and artificial intelligence will automatically write an email for us. We no longer need to write materials and reports ourselves, we only need to give it enough information, and artificial intelligence will automatically help us organize the text, and the writing is excellent. We no longer need to spend a lot of energy on writing plans and documents, but only need to think and come up with various creative ideas, themes, and ideas... Then tell these ideas to artificial intelligence, and it can automatically help us get all these tedious matters done. Even thinking itself can be assisted by artificial intelligence - you can give it a topic and let it help you find materials, cases, and references; or let it help you brainstorm and provide a variety of ideas for your reference...
- 3.4 Private Consultants
In the future, each of us may have our own doctor, lawyer, financial manager, etc. No matter what we want to ask, we can just ask the artificial intelligence, and it will give the latest, most comprehensive and most accurate answers. Of course, the current artificial intelligence is far from being able to do all of the above. But it is very likely that we will not be far away from this day.
Summary in plain language: The future prospects of ChatGPT may completely change the way we obtain information, read and study, write, and get professional advice, allowing us to enter a more intelligent, efficient, and personalized search 3.0 era, thereby greatly improving our life and work efficiency.
4. Train your own AI
We may truly usher in an era of "artificial intelligence assistants."
Among these applications, the one that interests me the most is actually fine-tuning. What does it mean? It means that on the basis of artificial intelligence itself, you feed it some unique and exclusive materials to train it into your own artificial intelligence, so that it can answer some more targeted questions and meet your unique needs.
Let me give you some examples.
Someone abroad shared an example: she uploaded her childhood diary to an artificial intelligence model, and let the program learn her thoughts, experiences and words, thus creating a "childhood self". Then, she talked to this "childhood self". She asked her about the problems she encountered, the confusion she had, and the choices she faced, and looked for answers from her heart. This is a very interesting example, and it is also very unforgettable to me.
Similarly, if you input all the works of a writer, perhaps with his life story and experiences, this AI can answer some relevant questions in the writer's voice. Someone abroad made such an attempt, inputting the works of Taleb, the author of "Antifragile", into ChatGPT, and asked it to answer questions seriously, forcing Taleb himself to come out and correct it.
So, if we feed Dream of the Red Chamber and various historical materials from the Ming and Qing dynasties to a more mature artificial intelligence, is it possible for it to continue writing Dream of the Red Chamber? I think it is possible.
Through fine-tuning, we can create an AI that belongs to us alone. Just feed it all our notes, materials generated at work, daily thoughts and diaries. In this case, it is you, and it is even closer to you than you are. Because it can faithfully retain many memories that you may have forgotten, and improve its prediction model based on a set of algorithms to make choices that are more in line with your heart and needs. In other words, this is an "inner brain" that belongs to us. When we have doubts, when we need to search for information, when we need to analyze problems and make judgments, we can refer to its opinions.
In plain language, we can fine-tune the technology to train our own AI assistant and apply it to various situations, such as self-exploration, content creation, or decision-making reference. This will become an indispensable “inner brain” in our lives, helping us process information, solve problems, and make decisions in a whole new way.
5. Disadvantages and Problems of ChatGPT
ChatGPT is not absolutely perfect. In fact, as of now, ChatGPT still has several problems that need to be solved.
- 5.1 Error Rate
The most serious problem is that in some areas, ChatGPT has a high error rate. This may be related to the fact that ChatGPT's training corpus is mainly English materials. If you ask questions in English, the accuracy of the answers will increase, but there will still be errors.
In addition, ChatGPT often answers questions in a very confident and confident tone. This means that if you are a layman, you may not be able to tell the accuracy of the answers it gives. Even for experts, it is difficult to verify some trivial details.
- 5.2 Moral and ethical issues
ChatGPT is an AI program that will only answer questions faithfully according to its internal predictive model. Therefore, if you ask it leading questions, you may get answers that deviate from ethics. This may cause unnecessary negative reactions.
Although the official has been constantly fixing vulnerabilities, it is hard to prevent them. Moreover, the more you fix, the fewer scenarios this tool can be used in. This may be a balance that needs to be considered.
- 5.3 The level of information intake
The last and very important question is the level of information intake. ChatGPT cannot generate information out of thin air. All its knowledge can only come from the corpus it is fed. Then, it is obvious that the answers it can provide and the content it can output will not exceed the average level of these corpora. You can't expect it to take in a lot of gag posts and ask it to write "The Three-Body Problem". And because there is a lot of low-quality information on the Internet, this leads to: the more information ChatGPT takes in, the lower its overall level may be. Unfortunately, we still don't have a reliable method to objectively judge the quality of content. Highly cited papers may be wrong, famous scholars may be spammed, and KOLs with many followers may be marketing accounts. If we rely on data such as users' likes or reposts to judge the quality of content, common sense and experience tell us: this will inevitably bring a lot of sensational and eye-catching low-quality content, which will only pollute the pool more seriously.
Moreover, as a large number of tools and services begin to use tools such as ChatGPT, a large number of articles generated by AI will flood the Internet. This will lead to "self-production and self-sales". Once AI can only be fed with corpus produced by other AI, its level will encounter a barrier that is difficult to break.
Some studies have suggested that within 5-10 years, the content on the Internet may be exhausted. By then, artificial intelligence will be left with nothing to do but regurgitate the information it has produced.
Summary: Although ChatGPT has great potential, it still has some key problems, including high error rate, ethical issues, and the quality of its output content is limited by the quality of input information. These problems remind us that despite the rapid development of artificial intelligence, we still need to pay attention to and deal with the problems it may bring.
6.What jobs might ChatGPT replace?
It seems that every time we talk about artificial intelligence, someone will ask this question. However, the experts' answer is still relatively optimistic. As of now, and even in the near future mentioned above, AIGC tools such as ChatGPT will not replace anything - it will only improve our efficiency.
The reason is simple:The positioning of AIGC is not a "function" but a "tool".
What is a function? For example, if there is an artificial intelligence that helps you read X-rays, then this is a function, and it may replace some of the functions of doctors; but the functions of AIGC are searching, reading, and writing, which are basic skills required by all professions. It is essentially auxiliary and is an auxiliary tool.
What can AIGC do for us? We no longer need to write a bunch of repetitive lines for each NPC. Instead, we can design the background of the NPC and let AIGC automatically generate their lines, and even generate tasks and plots. We no longer need to set up the personality of each character in the novel. We can first make a background framework, and then let AIGC generate a large number of characters, and we can choose from them and polish them...
Therefore, with the existence of AIGC, we can imagine the following future: information search will be more accurate, reading efficiency will be higher, there will be more channels for reference and comparison of professional advice, game plots and tasks will be richer, novels and TV series will be more diversified, and news can be pushed to us in real time...
All creative professions should be happy about this because the biggest obstacle that plagues them - boring, tedious, labor-intensive operational work - can be replaced by AIGC.
Summary in plain words: ChatGPT and other automated information generation and consumption (AIGC) tools are more like tools to improve our efficiency rather than functional replacements. They can help us handle tedious and labor-intensive work, allowing us to have more energy to focus on more innovative and creative tasks.
7. How to use ChatGPT
1) Those who have Mofa Hoof can directly visit the official website to use it;
2) If you don’t have a hoof, you can use a domestic mirror site;
8. Conclusion
So, what is the most needed and perhaps the most important capability when it comes to AIGC tools such as ChatGPT?
—— is the ability to know what you want and to express your ideas clearly.
Being able to ask a good question is sometimes more important than being able to solve a problem. The emergence of ChatGPT makes this point even clearer. Your clear understanding of yourself, your unique insights into the outside world, your insight into the essence of things, and your ability to communicate and express them. These four points will be our weapons to be more relaxed and at ease when facing a world driven by artificial intelligence.
In the future, using artificial intelligence (AI) will become a must-have skill because the development of AI will create a huge efficiency gap in many fields. In some fields, the output efficiency of people who are good at using AI and those who do not understand AI may differ by 10 times.
Therefore, we must attach importance to and actively learn AI in order to better adapt to future development.