Get ready for new technology
The coffee shop in Beijing is the best place to learn about the current domestic trend. People sitting in the coffee shop chat from the blockchain to the metaverse, and then to today’s AIGC. Now, almost every coffee shop within the Fifth Ring Road has guests gathering to discuss AI.
This wave of technology is coming fast, and AIGC (that is, generative AI, AI Generated Content) is more realistic and has more room for implementation than the Metaverse. Among them, the emergence of ChatGPT is very important, so that everyone can know, feel and operate the application, which is an important boost for AIGC to quickly sweep all walks of life at home and abroad.
Another equally popular concept is that of "megamodels". Lu Qi, former president of Baidu and founder and CEO of Qiji Chuangtan, mentioned in a speech a few days ago that OpenAI brought the era of large models, and the future will be an era of "models everywhere".
According to the statistics of Minsheng Securities, nearly 30 large-scale models have been released in the middle of April, becoming a new outlet for entrepreneurship. Zhang Yong, chairman and CEO of Alibaba Group, said: "Facing the AI era, all products are worth redoing with a large model."
Let us turn our attention back to the film and television industry. At the online audio-visual conference that ended at the end of March, almost everyone was talking about AIGC. ——Compared with the competition between long and short videos two years ago, this year's platform stories are obviously all about AI.
But how should it be combined? It seems that there is no clear answer yet. Sun Zhonghuai, vice president of Tencent, once mentioned, "At present, it seems that ChatGPT should be applied to the basic editing of TV drama scripts, so as to save labor costs, but what other possibilities are there?"
This is why, compared with games, advertising, IT and other industries where some positions have been replaced by AI, the film and television industry has not been greatly affected. Although many people are paying attention, discussing and researching, but it is not really implemented and put into application. not much.
Dumou (ID: Domoredumou) consulted some practitioners in the film and television industry, and got similar answers—they don’t think AIGC will fully replace creators in the future, because “inspiration and creativity cannot be replaced by machines”, There are not many front-line practitioners who are actually using it. Obviously, entrepreneurs in this wave of AI do not think so. In the stories they describe, new technologies will be the ultimate answer to all problems.
Therefore, in the future, we will continue to have dialogues with entrepreneurs who have entered this track, trying to find the real connection between AI and the film and television industry, as well as games, music, short videos, etc. To what extent is the industry we are concerned about being or will be changed by AI.
Our guest in this issue is Yin Bohao, the founder of Monkey Unlimited. Monkey Unlimited is an AI company established in August last year, focusing on model tuning. Their first investment at the start-up came from Lu Qi, known as the "Chinese AI Evangelist". According to his recollection, it took only ten minutes for the investment to be completed.
Yin Bohao started to engage in related work in 2015. He has been paying attention to the development of the AI field and looking for possible entrepreneurial opportunities from it. What really prompted him to join this wave of entrepreneurship is the arrival of the technological inflection point he judged, that is, after many iterations Post-mature ChatGPT4.
Currently, Monkey Unlimited has only five regular employees, all of whom work online. He joked that they are the next generation of "digital office models".
The following is the transcript of the conversation:
AIGC, just the beginning?
Poison Eye: In your judgment, what is the biggest change brought about by the big model?
Yin Bohao: Many people think that AI is the end, but in fact AI is a foretaste, because the change it brings is that all information in the human world can be processed in a new way.
The core change is that the way AI uses data has undergone great changes compared to the past. For example, in the past 1.0 era, such as Tencent conferences and WeChat, they did not generate information themselves, but only carried information and had no information processing capabilities; In the 2.0 era, it may be artificial modeling, or weak AI-assisted models, such as Excel, CRMERT, HR human resources management software, etc. Many rules and processes are driven by humans, and it is still very difficult for humans to use data in this process , requires a lot of cost, either human annotation, or human design process.
What ChatGPT3 does is almost continuous learning. Put a lot of data into this model, and it will learn automatically. Iterating to 3.5 and 4 is actually a process of reinforcement learning. Compared with the previous two generations, either there is no information processing ability at all, or there is some information processing ability but a lot of human participation. Now it is an almost fully automatic process.
Poison Eye: So what it changes is the way the human world processes information?
Yin Bohao: In the past, when we wanted to write a script, we needed to read a lot of scripts and then summarize the rules. For example, the model of "Journey of Heroes" is actually a process of artificial modeling. If there are enough texts, we can get all the scripts in the world. Then I can write a script like this.
Similarly, if I want a script in the style of Jiang Wen, what style is Jiang Wen? In the past, we needed to watch all his films to understand and summarize. If there was an exclusive large model of Jiang Wen, we could ask AI to write a "Jiang Wen style" story.
Poison Eye: It is equivalent to AI replacing the process of human summing up the law, and it can also reuse this law.
Yin Bohao: Yes, this is a very important feature. The image is the same. If you cut out every frame in all Jiang Wen's movies to build a model, it can learn Jiang Wen's image style. This is the era of large models, many people do not see these opportunities, think it is just to help us write something.
Poison Eye: What are the cases we have put into use so far?
Yin Bohao: We are serving a brand of disposable e-cigarettes. They have 80,000 historical emails, which are the opinions of some customers. After building the model, it will automatically learn and disassemble the content of these emails, and complete the work of email reply.
Poison Eye: It sounds like TO B’s business logic.
Yin Bohao: The core problem we solve is from data to intelligence. At present, it is still TO Big B (serving large enterprises). The essence is that you give me data, and I will give you a model, and let this model successfully operate within the enterprise. At present, we have reached a cooperation with Haier and produced an innovative design platform for them.
Of course, if we discuss personal data to personal intelligence, it is digital life. Using your historical chat records with everyone, you can train your dialogue model, and the articles you have written in history can also be trained as your own writing model. We are ultimately a company that drives digital life with large models. Next we will also work on everyone's digital life, but it will take some time.
Poison Eye: Glow, a relatively popular application some time ago, is also the idea of customizing the model.
Yin Bohao: Yes, we also competed with Glow for customers, and we won the order. Our technical logic is exactly the same, that is, general model + private data = private model, but the business logic is different.
Poison Eye: Which type of companies do you currently serve or contact? We are more curious about what industry people will be more sensitive about this matter.
Yin Bohao: There are all of them. At this stage, if you pull someone on the street, there is no one who does not stop chatting. I feel that the length of response time does not distinguish between fields, and Internet companies are not in a hurry. People who make quick decisions always make fast decisions, and people who make slow decisions always make slow decisions.
However, there are indeed no customers in the film and television industry, and they may not know us yet.
Poison Eye: In your opinion, where will the combination of AI and film and television be?
Yin Bohao: Everyone's imagination is still too lacking. They just use AI as an auxiliary tool. "It used to take me a day to shoot a thing, but now it only takes 10 minutes." ——Imagination ends here.
But the real concern is that new mediums can produce art forms. The medium is an extension of people, and AI2.0 itself is a new medium. If you train a digital life, is this an art? Is it art to train a robot that can chat with me every day? In my opinion, the boundaries of art should be expanded by media.
Poison Eye: So what can really bring about change is the digital life?
Yin Bohao: The core is digital life. A chatbot is a kind of art, and if I create a new world that already has a basic setting, with a bunch of agents in it, then it is a new art form in itself.
Poison Eye: It sounds more like an extended form of the game.
Yin Bohao: Yes, so games are a bigger market than film and television, and the boundaries between film and television and games are gradually blurring. This may be the next generation of art. Perhaps in the future, the movie will become the leading film of the game, and a new world will be designed first, and everyone in this world will have a different plot.
Poison Eye: Will the traditional form of film and television be completely subverted?
Yin Bohao: No, it’s like there are still people watching dramas now.
What AI cannot replace is...
Poison Eye: As you just mentioned, if you import all the information and materials of a certain person to generate an exclusive model, copyright, legal and even ethical issues will inevitably arise. The Measures for the Administration of Generative Artificial Intelligence Services (Draft for Comments)” has been open for comments. How do you view the emergence of these problems?
Yin Bohao: This is not an issue we need to consider at this stage. When technology has advanced to a certain level, someone will naturally summarize these issues. When I participated in the AI closed-door meeting before, I met a documentary director who has won two Emmy Awards. He also came to us to talk about cooperation. If you don't take the initiative to use your own data to train this AI, someone else will steal your data to train AI.
Poison Eye: I chatted with a film director a few days ago, and he said that if there is a very mature AI that can be used by him, he can complete his own movie with the AI, which will greatly improve the efficiency.
Yin Bohao: AI assisting humans is already a very deterministic thing, so that I think it is still worth talking about now? Hasn't it already happened?
Poison Eye: How long does it take to generate an exclusive large model?
Yin Bohao: Very fast. Everyone has a misunderstanding about AI, thinking that the training time is very long. Of course, because AI learns all the knowledge in the world, it needs to learn biology, physics, history, Lu Xun's speech, and Marquez's speech. Of course, it will take a long time. If you only learn one person, one day is enough.
Poison Eye: Now we still talk a lot about Wenshengwen, Wenshengtu and Tushengtu. Will entering the 3D mode have a greater impact on the film and television industry?
Yin Bohao: Yes, because 3D content is very scarce, it costs tens of thousands of dollars to build a 3D content model. For example, what NeRF is doing is that I need about 50 pictures to scan a circle to generate a 3D model, but if combined with AI, one picture may be enough. I judge that this will happen within six months to a year.
Poison Eye: What about generating video from text?
Yin Bohao: Also half a year to a year. As I just said, people who move fast have already started when the technology is very immature, and people who are slow may never start.
Poison Eye: Nowadays, when many film and television and animation companies manage digital assets and access related plug-ins, they will encounter the situation that each department is not unified. Can the large model solve this problem and help these companies realize "asset unification"?
Yin Bohao: The large model itself is a compression method. In the past tens of thousands of pictures were tens of thousands of files, but the large model can compress tens of thousands of files into one file and decompress it losslessly at the same time. The large model is compressed and decompressed. process.
Poison Eye: At the same time, it can also classify independently.
Yin Bohao: Yes, maybe 10,000 pictures of each of the 10 characters are all compressed into one model, which can be flexibly decompressed to produce 10 characters. You can understand a large model as an executable compressed package. I think a large model has three core features, self-training, compression, and generalization. You just mentioned compression.
Poison Eye: Nowadays, many people generally feel anxious when facing the new wave of technology. If large models can really replace many jobs, what will happen in the future?
Yin Bohao: First of all, everyone has a base income (basic salary), which will definitely happen, because many people will be replaced.
Another dimension is related to each individual. For example, I used to drive a horse-drawn carriage, but now I drive a car, so new technologies will also bring new jobs. In the past, many people had to write code line by line, so writing code was very expensive. Now it is very easy to write code. Everyone can write code and create their own things. The total amount of code in this world will be more, and there will be more job opportunities. Also more.
The third is that there will be new jobs, including new jobs in our company, data cleaning, model training, etc. These are all new jobs.
Poison Eye: The last question, as content creators, we think that the part that AI can never replace is human inspiration, emotion and innovative ideas. Do you agree with this point of view?
Yin Bohao: They can only brainwash themselves to say "I am Jiang Wen". I think that AI may only be able to reach the level of a person who has written scripts for three years, but it will continue to break through in the future. It may be able to write other than Jiang Wen's, and it can imitate Jiang Wen.
Poison Eye: So if we put aside the perspective of content creators, have you ever considered what is the part that AI can never replace?
Yin Bohao: It was quite difficult.
Poison Eye: For example, if we train a large model that can train a large model, can you also be replaced in the future?
Yin Bohao: That’s also possible, all of us must be prepared to get base income, it’s a matter of time.
To experience the most cutting-edge AI tools, you can enter AI Tool Cool: allin.aigcgeek.com has collected 500+ top-notch and cutting-edge AI tools and products at home and abroad by category. In the AI era, it is enough to start with this Tool Cool.
For more ChatGPT, Midjourney dry goods skills, welcome to the official account of the same name [AI Hurricane]