“An astonishing fact is that almost all companies, whether they are in gaming, live streaming e-commerce, new consumption, education and training, fitness, podcasting, short videos, etc., are discussing a common topic: how will AIGC represented by ChatGPT enter and change our industry?”
The fact that has already happened is that while major companies are downsizing, they are also betting on AIGC. Even startup companies are downsizing, which seems to warn each of us to “recognize the situation and give up illusions.”
Besides improving efficiency, what paradigm shifts will this wave of AI technology bring to the content field?
Chaos held the “One” Thinking Innovation Carnival at the InterContinental Hotel in Dianshui Lake, Shanghai. With an entire island and a two-day, two-night shared learning venue, 2,000 chaotic students enthusiastically participated! An invitation was extended to an all-star lineup of AI instructors.
- Messari Interpreting Stacks 2023 Q2 data, overall performance is good.
- 8 Alpha projects worth paying attention to $PENDLE, $RDP, $WLD…
- Worldcoin Controversies and Criticisms
During the roundtable discussion at the Chaos “One” Thinking Innovation Carnival, Bai Yunfei, Brother Egg, and Teacher Zhang Xiaonan discussed the question of “How does AIGC change the content creation industry?”
Below are the shared notes:
01 Why is the content industry the first to be affected?
Zhang Xiaonan: Teacher Yunfei, you are engaged in graphic content entrepreneurship. In April of this year, you achieved cost reduction and efficiency improvement due to the application of ChatGPT. Many people have known about ChatGPT for a long time but have always been unsure how to apply it. How were you able to apply this technology so effectively?
Bai Yunfei: There was a reason behind this decision. In December of last year, a foreign friend of mine messaged me, saying that a tool called ChatGPT would be of great help to our business and asked me to give it a try. After trying it out, I found that this tool did indeed greatly benefit our business. It wasn’t until February of this year, after the model had been running stably for more than two months, that the production results were truly the same as those produced manually, or even better. That’s when I felt the timing was right.
Zhang Xiaonan: ChatGPT has indeed brought about changes for many companies, and it is often the content team, video team, and design team that experience the changes first.
Brother Egg has a deep connection with Chaos. He started getting in touch with Chaos seven years ago and has always been a loyal student and fan of Chaos. At that time, he set a goal for himself: “Chaos has a great field, Chaos has great courses, one day I will definitely give a lecture at Chaos.” Today, seven years later, he has achieved it. Two weeks ago, his lecture on content entrepreneurship and short video entrepreneurship was launched on the Chaos platform. He has an account called “Danjie Entrepreneurship,” and in reality, he is doing MCN and has already created many matrix accounts.
Let’s talk about how the application of AI technology has changed things for you, Dan.
Dan: We mainly use AI to draw comics and create comic-style videos. You can follow an account called “Smart Two Pillars” to see the content produced entirely by AI, including writing, drawing, and editing.
Personally, I think AI is quite down-to-earth. Many people may think AI is just a large language model, but in my opinion, AI drawing software is like a new era of Photoshop.
For example, ten years ago, if you learned Photoshop, you could do graphic design. But now, if you learn Stable (stable diffusion), you can use AI to create comic videos without any drawing skills. In the past, it would take at least four people, one writer, two artists, and one editor, to produce a comic video in a day and a half. But now, with Stable, one person without any drawing skills can create a video in just four hours. So, efficiency has increased and costs have decreased.
As we all know, MCN bosses often have conflicts with popular influencers because real people are unique products. They can do anything when they don’t have any fans, but once they have fans, they can’t do anything. So, when we first started making comic videos, it was to address the standardization issue and avoid dealing with real people. But then we encountered cost issues when creating comic IPs. It’s quite expensive to have four people working on one account. That’s why we started seriously researching AI.
We have created a video that we can play for you.
I believe you may think this video is quite low-quality, and that’s right. Short videos themselves have subverted long videos in a low-end way. In the past, you had to use a DSLR or even a professional camera to make a video, but would anyone watch it? However, most short video creators can shoot videos with just a smartphone, and that’s the low-end subversion.
The reason we are making such low-quality videos is not because we can’t create better comics. Actually, AI can create more stunning comics. But we don’t think it’s necessary at the moment. As long as we can find the most cost-effective style and create comic videos quickly and at a low cost, that’s enough. This seemingly low-quality video has received 100,000 likes on Douyin.
Zhang Xiaonan: 100,000 likes and nearly tens of millions of views. Low-end subversion is also one of the models we have been learning at Chaos.
I totally agree with what Dan said. We also make videos every day, and I’ve found that sometimes when you sit upright and carefully adjust the lighting and camera position, the video effect may not be that great. On the other hand, when you casually shoot something, the effect is often very good.
By the way, Yunfei does graphics and text, Dan does videos, and I do live-streaming e-commerce. AI has been very helpful for all of us. You mentioned earlier that AI has helped you improve efficiency. Do you think there are any other possibilities or areas for expansion?
Dan: Actually, we have tried using ChatGPT to write scripts. However, we found that pure ChatGPT script writing is not realistic in the short video industry because it sounds too rigid. Whether it’s version 3.5 or 4.0, it can help with topic optimization or finding ideas. If it’s not limited to Douyin, it can generate content with just one click. For example, it can create a Little Red Book-style copy for an ordinary article, and that works perfectly fine.
However, if you use ChatGPT to directly write scripts for short videos, it may not be grounded because I believe that the essence of content platforms is actually emotional platforms. Good content is not about serious matters, but about evoking emotions. Because ChatGPT cannot evoke emotions, it can only provide content value, which is an essential aspect that humans cannot do without.
02 Will the extensive use of AIGC lead to the homogenization of content?
Zhang Xiaonan: If we all use such tools to produce content, will there be a situation where there is a large amount of homogeneous content due to everyone quickly using AI tools to create accounts, whether they are graphic or video accounts, and causing user aesthetic fatigue?
Bai Yunfei: Actually, my answer deep down is no. Why? Because every person trains AI in a different way, and each person’s prompts and parameter adjustments are different. In this case, the resulting generated content will have significant differences, making the possibility of homogenization relatively small.
I firmly believe in one thing, that is, when creating content, it must have its own unique style, like an IP. How to better collaborate between humans and AI and make the things generated by ChatGPT more personalized? At this point, you will find that the possibility of true homogenization is very low. This is my personal opinion.
Zhang Xiaonan: Can you give a specific example using a scenario or case from your work?
Bai Yunfei: Let me give an example from our Zhihu business. The Zhihu team created a booklet with over 80 pages, which mainly includes various structures of product copywriting. We used this material to train ChatGPT, and in addition to that, we also optimized some details, such as studying which wording can encourage customers to place orders more easily. We study user psychology and combine many things to ultimately form a high-quality model corpus. After that, you can continue to feed and train it, and the effect of the content generated by GPT is very obvious.
Dan Ge: When it comes to the issue of content homogenization, we first need to clarify what content is. I think most companies are not producing content, but rather chasing money.
For example, Picasso is a painter, but the ones who really make money are those who imitate Picasso. Are these people called painters? Actually, they are not called painters. They just earn money by imitating Picasso, and the medium is a painting. Whether it is Zhihu, Xiaohongshu, or short videos, the replication of content is a consistent business tactic. Replicating content can effectively increase followers and make money, nothing more.
What can be done about content homogenization? On Douyin, most e-commerce sales are done by repetitive copywriters. If a copy has a high conversion rate and ROI, different anchors should be used to promote this copy. If you want to discuss how to be an elegant cultured person, then don’t talk about how to make money. If you want to talk about making money, then don’t talk about how to be a cultured person.
When I participated in the Douyin 2D Manga Contest before, the people below were all content creators, comic artists. They went on stage and talked about their dreams of making a Chinese blockbuster, changing the comic environment in China, and creating the first Chinese Disney. When I went on stage, I talked about wanting to earn money through advertising.
So regardless of whether you treat AI as a large language model, a future trend, or a tool like Photoshop, the core question is, what is AI for you? Is it a tool to make money or to pursue something higher?
In fact, most AIs are only suitable for companies like BAT and ByteDance. We should talk about practical things, how an individual, a small company, can use it as a tool and make money.
Zhang Xiaonan: I fully agree with what you just said. Putting knowledge into practice is more important than anything else.
Many people want to make money. Why can Brother Dan immediately use AI to empower himself and make money, while most of us have not been able to do so? When you first came into contact with AI, what made you believe that it would work and that you should use it?
Brother Dan: I think whenever a new thing comes out, the first wave of people who use it are always the ones who benefit from it. The reason why I studied this matter at the first opportunity was because I was truly afraid that someone would use AI to create a bunch of comics more efficiently and at a lower cost than us, and ultimately overthrow us. After using it for a while, I found that AI can indeed help me reduce costs and increase efficiency.
The next step I thought about was how to continue making money with it. I think selling AI courses is the most profitable thing to do at this stage.
Zhang Xiaonan: History always repeats itself in certain forms. Some time ago, everyone knew that the most profitable thing was teaching people how to do IT, branding, and short videos. Now, all these people are teaching people how to do AI. It’s like what we said about the gold rush. You may not find gold, but you can definitely make money by selling shovels and jeans. That’s just how life is.
Let me ask Yun Fei again. Many people say that the dividend of text and images is gradually declining, while videos and live streaming still have a lot of potential. Why do you continue to focus on text and images? What is your reason for sticking to it?
Bai Yunfei: Currently, content entrepreneurship is a hot topic, and many people are trying it. However, we have found that in order to stand out among many competitors, the probability is relatively small. The key to content entrepreneurship lies in the content itself. High-quality content can attract more users. What is the first thought that comes to mind when people think about content? Should they start a live stream or open a short video account? Maybe they will become popular someday? But after trying it ourselves, we found that the majority of content platforms currently use a centralized streaming mechanism, which means that the chances of an ordinary person becoming popular are relatively small.
At this time, we also have a cake – text and images, which gradually became less popular and people started chasing short videos instead. This direction gradually became a relatively untapped market, so we chose to focus on text and images. So far, we have made a good amount of money.
Zhang Xiaonan: How many people are there in your team now?
Bai Yunfei: Our company has a total of 43 people, and the content team has 21 people. Among them, the Zhihu section now only has 4 people after being assisted and optimized by GPT. We have always believed in one thing, that in content entrepreneurship, content production and content creation are two different concepts.
If you really want to make money through content, I think the focus should be on content production and turn it into a standardized process, just like producing goods.
Some people think that they need to inspire their own inspiration and create unique videos. But if you approach it with this mindset, many accounts won’t be able to succeed.
03 Which AI tools have actually been used in work?
Zhang Xiaonan: In the field of content entrepreneurship, what are the AI tools that empower your accounts, content, or company? How do these tools help you in your work?
Bai Yunfei: We have been doing text and images, so we commonly use two tools:
The first one is ChatGPT4.0, which mainly helps us produce content. We have 4 operations in our Zhihu business, and each person produces 40-50 pieces of content per day. These contents are all produced according to our high standards. These standards are pre-set in the model, which can greatly improve efficiency.
The second tool is Midjourney. When we express something, we need to add images to an article. In the past, after writing the article, our colleagues would search for some images online and choose the ones that best match our expectations, and then incorporate them into the article. Now, we can let Midjourney generate images that better express our ideas according to our intentions. This way, articles with higher standards and images that have personalized expression intentions can better express our ideas, and the traffic we get on the platform is very good.
Zhang Xiaonan: Da Ge started with audio, and everyone can feel his strong IP attribute because you are an IP yourself. If you switch to comics now, it will to some extent weaken your IP characteristics, but AI can empower more comic attributes. How do you consider this?
Da Ge: First of all, weakening the IP is because I don’t want to still be making videos and live streaming when I’m 50 years old, relying mainly on physical strength. Whether your body is strong or not determines the revenue of your company.
So it is a process of weakening the IP.
The AI tool our company mainly uses is Stable.
Zhang Xiaonan: I have found that the people who can truly become IPs are able to use vivid language to explain a problem that you have been thinking about for a long time but couldn’t figure out in just one sentence.
For those of you who have achieved results, do you feel any pressure? What is the development speed of your peers in the field of AI?
Egg Brother: The reality is that most of our peers are only pursuing short-term speed, so they use AI to create a novel, but it used to be an audio novel, and now it is accompanied by comics, and they consider it an AI work. But I think this is too low-end and cannot subvert anything, so we need to raise it a little bit based on the low-end.
Currently, our peers are mainly focused on selling AI courses, so I would like to remind you all that there are many courses being sold now, and you can listen and learn more, but that cannot become the main direction of the company.
When it is truly used by the majority of people, it should be optimized into a more user-friendly mode by more companies. That is what the general public will use. Currently, the B-end is nothing more than transforming and making money in this area.
Zhang Xiaonan: Do you think AI tools are really that difficult to use? Or are there any barriers?
Egg Brother: Actually, there are no barriers. You don’t need education or any hard requirements, as long as you are willing to learn. So, in my opinion, AI is like a current version of Photoshop, just like the spinning jenny in the past. With the spinning jenny, people can weave fabric faster, and with Photoshop, people can do graphic design. It’s essentially the same.
04 Two key issues to solve when using AIGC effectively
Zhang Xiaonan: What do you think are the two key issues that companies need to solve when using AI to generate content?
Egg Brother: I think: firstly, you need to have content capability. For example, if you develop a particularly powerful pair of football shoes that can improve accuracy by 10% for anyone who wears them, but if you let certain teams wear them, they still won’t qualify for the World Cup. This tool should be used by people with a certain foundation, so that they can achieve incremental improvements. It is not something that provides immediate benefits, but rather something that enhances existing strengths.
Secondly, I think it is the ability to learn. AI is updating and iterating at a very fast pace now, and if you don’t have the ability to continuously learn and update, you will fall behind.
Bai Yunfei: Because our large-scale language models need to be trained, without training, they will always be generic. For example, if you find that when Zhang San and Li Si ask GPT the same question, the produced results and text will have slight differences, but it is only a slight difference, and they still tend to be homogeneous. So training is still necessary.
As an example of what we are doing, the most crucial issue is to classify the content. For example, among the 15 types of product promotion we have done before, we classify them and determine how to start each type, how to end it, where to insert persuasive words that can stimulate consumers to place orders, and so on. We will fully organize the entire result.
You can think of it as a template, and each article will have a template. This template may be 700 or 800 words long, and it is a formula. After extracting this thing, you can import it into ChatGPT as a corpus for training. The final effect will be much better than directly inputting “Please help me generate an article with product placement” in the box. The effect can be several tens of times better at least.
Zhang Xiaonan: I completely agree with what Yun Fei and Dan Ge just said. I also have a very real feeling in my work scenario. Once during a live broadcast, I had two courses, one was a science course and the other was a critical thinking course. In chaos, critical thinking course is in great demand, but for children, critical thinking or science courses seem to be not as necessary as Chinese, mathematics, and English. For this course, I read two books: “The Story of Science” and “The History of Science”. After reading these two books, I kept having conversations with ChatGPT, asking it to tell me how to sell a science course well. At the beginning, the content it generated was really rubbish.
Why is learning to ask questions important? In the book, I saw a lot of content, absorbed it continuously, and then iterated and had conversations with ChatGPT. During the process, I found that ChatGPT became more and more powerful. As a result, both the science course and the critical thinking course sold very well. ChatGPT did help me a lot.
A critical thinking course, which is not in great demand, sold millions in that live broadcast. I think ChatGPT has been quite helpful to me because it lets everyone know at a deeper level why critical thinking is so important for children.
Dan Ge: Let me add one thing. You just mentioned that the critical thinking course is not in great demand, while Chinese, mathematics, and English are in demand. In fact, I think the next trend and opportunity may lie in the design of courses.
I believe that in another 5 or 10 years, the education received by the current generation of children should prioritize critical thinking. If they can ask good questions, a lot of knowledge can be obtained through questioning. So I think the first wave of opportunity is to sell AI courses, and the second wave of opportunity is to tell everyone that you should let your children learn critical thinking in the AI environment.
Zhang Xiaonan: I completely agree. AI won’t subvert what kind of people? Those who have critical thinking and innovation abilities.
Just now, I also want to ask Yun Fei the same question. What do you think are the two core problems that need to be solved when using AI in the content field?
Bai Yunfei: The first point, I think, is active learning, as Dan Ge just mentioned. A teacher recently conducted a survey, and the proportion of students who have used related AI tools on-site, such as ChatGPT and Midjourney, is very small. In this case, I recommend everyone to experience it, and after experiencing it, understand the basic principles. Based on this foundation, I think it should not be difficult to use it.
Secondly, it is still about data. Because artificial intelligence training always has a prerequisite – high-quality data. If you feed garbage into it, what comes out in the end is garbage. If you feed high-quality corpora and data into it, what comes out in the end is a very excellent language model.
The key is how to find more excellent data sources. Many people should know better what is relatively high-quality in their own company, their own business, or their own life and work, and then extract these high-quality data to feed in. The effect will be much better.
I think the above two points are what everyone urgently needs to do based on the current situation, which can be regarded as a proactive attitude towards embracing AI.
05 Regardless of the era, IP is of paramount importance
Zhang Xiaonan: The formation of IP is often inherent in the person itself. When it comes to individuals, people will mention character design and the style of the genre. Are these still important now?
Egg Brother: I think character design is still important. Our comic account has a decent advertising ability, but the conversion rate of selling products is particularly low. The problem with non-human IP is that it lacks user stickiness, and users’ trust will be very low. When selling the same product, using the same popular script and comic IP is not as good as selling it with a real person.
Character design and style have always been important in any era, including previous newspapers, graphics and texts, public accounts, news, and now short videos, etc.
Bai Yunfei: The anxiety we just mentioned is that in the future, with the popularization of cost reduction and efficiency improvement, many platforms may be flooded with homogeneous content. In this environment, I believe that accounts with strong IP and unique character design will stand out.
For example, suppose you come across two accounts now, one is Egg Brother and the other is a virtual character. Obviously, everyone here is more familiar with and has more trust in “Egg Brother” than in the virtual character. This familiarity and trust will naturally lead people to prefer to communicate with “Egg Brother,” thereby improving the efficiency of content monetization.
Therefore, a strong IP and character design can win more trust, and trust further promotes transactions. This is the most important goal that we, as content creators, should pursue.
Zhang Xiaonan: I think Yunfei mentioned a key keyword. Whether it is short videos, graphics, or live-streamed e-commerce, trust is definitely at the core. If your account and content can completely break through in terms of trust, your monetization efficiency and conversion efficiency will definitely be high.
I would like to give a small suggestion to all partners who want to enter this field. IP is difficult to cultivate and train. It is more about people who naturally possess certain attributes and characteristics, just like Egg Brother, whose personal characteristics are very obvious.
I would like to ask if you have encountered any difficulties and challenges in using AI tools?
Egg Brother: The first difficulty and challenge is the lack of technical talent. When AI is upgraded and iterated, we are slower than technical teams and need to find other teams to assist us in completing the upgrade, so our learning costs are twice as high as theirs.
In addition, the real effect brought to us by AI is only the creation of hundreds of thousands of accounts. But I think in the future, AI can completely achieve one-click generation from text to images. So the biggest challenge for me is whether our industry is still worth existing.
Bai Yunfei: ChatGPT has a common problem, which is to talk nonsense in a serious manner. For example, if you give it a word that doesn’t exist at all, it will still explain it seriously. This is also a common problem of the model and cannot be solved in a short time. This is a challenge.
Another challenge is from the perspective of the public. For example, if I input a topic now and want it to immediately generate a usable article, to achieve this effect, there are actually many things behind it, such as the NLP-related basic knowledge mentioned by the previous experts at least need to be understood to some extent.
In the face of this scenario, if you input to AI, “I need a very high-quality Little Red Book note,” you will find that the effect is poor. Because your command is too general. So you need to figure out what style you need, what tone, even how many words, what kind of role, and even what kind of scene. After figuring out everything, the effect will be better.
Zhang Xiaonan: Speaking of content, let me also share a feeling with you. Content is not just for the sake of content. If you are for content, you won’t think about these techniques and methods. You will create like an artist, write like a writer.
If we treat it as a business model, there must be the word “monetization” behind the content. Before you do content production, you should think about the monetization path. Egg Brother’s monetization path is advertising, and Yunfei’s monetization path is mostly advertising, and then bringing in sales. When you have a clear understanding of the monetization path, what kind of content you want to create, including what tools to use, different companies have different choices.
We should be grateful for the Internet era, which allows us to take small steps and iterate quickly, and find opportunities for our company or enterprise in this process.
Like what you're reading? Subscribe to our top stories.
We will continue to update Gambling Chain; if you have any questions or suggestions, please contact us!