How has the controversial AIGC become a top player?

​Written by Qianshan

Since this year, AIGC has risen rapidly.

The so-called AIGC, or AI-Generated Content, refers to the use of artificial intelligence to generate content, which is considered to be the successor to professionally generated content (PGC) and user-generated content (UGC). The latest new way of content creation. The “AI painting” that sparked a heated discussion not long ago is a typical example of AIGC.

In this technology track, not only many established giants continue to make plans, but there have even been several unicorns valued at over one billion US dollars.

On October 18, Stability AI announced that it had received US$101 million in financing. It is the company behind Stable Diffusion, the most popular text-to-image generator in the open source world.

On October 19, Jasper, an AI startup focusing on text generation, announced the completion of a US$125 million Series A financing, with a valuation reaching US$1.5 billion.

In addition, according to statistics, 80% of AIGC concept stocks achieved profits in the first three quarters. The venture capital circle undoubtedly favors it, and it can be said that it has been in the limelight for a while.

Why is this concept, which is not new, suddenly so popular?

1. Why it became popular overnight

From the surface, everything seems to have started with a digital event held in Colorado, USA, in August this year. Artist competition. A painting titled "Space Opera" won the digital art category. As soon as the news was announced, it caused a lot of controversy.

##Théâtre D'opéra Spatial (Space Opera House)@Baidu Encyclopedia

The focus of the controversy lies in , this work was generated by game designer Jason Allen using the AI ​​drawing tool Midjourney. Although Allen polished it, this was obviously not enough to quell the controversy.

Discussions about "whether AI painting will replace painters" and "whether AI will lead to the death of art" heated up rapidly. After the incident fermented, more people suddenly discovered: It turns out that AI painting The level of development has reached an astonishing level.

In fact, as early as the beginning of last year, the image recognition system CLIP and language model DALL·E launched by OpneAI were a very important signal. After that, the field of AIGC began to bloom everywhere. In addition to AI painting, AI writing and AI programming tools also emerged one after another.

Since this year, AI auxiliary tools such as Disco Diffusion, Midjourney, and Stable Diffusion have become popular one after another, not to mention the continuous deepening of domestic and foreign manufacturers in this field:

Google Brain has successively launched AI painting tools "Imagen" and "Parti";
• Microsoft Research Asia has released the infinite visual generation model NUWA-Infinity;
• Github's AI programming artifact Copilot can generate sub-modules for developers to use based on massive open source codes;
• OpenAI has officially opened the program interface of DALL-E 2;...

The explosion of AIGC may seem sudden, but in fact it has also experienced a long dormant period, and "Space Opera" happened to be the lead that made it "famous and famous all over the world". In the final analysis, the blessing of multiple factors made AIGC famous this year.

First of all, the technical level has reached a certain level, especially the development of technologies such as generative diffusion models and multi-modal pre-training models have made significant progress, allowing AI to quickly generate data of different modalities. Content can be widely used for content generation, editing and auxiliary creation.

Then, market demand drives the evolution of content production methods. People's demand for content is growing rapidly, and the content industry also needs to be upgraded accordingly. PGC has a higher production threshold and guaranteed quality, but the production cycle is longer. UGC reduces production costs and centralization, but the quality is often uneven. AIGC takes into account both production capacity and quality to a certain extent and can be regarded as It is a new round of paradigm shift in the development process of content ecology.

Furthermore, the economic downturn has made technology more pragmatic. In the post-epidemic era, cost reduction and efficiency improvement have become the direction of many companies' efforts. Generative artificial intelligence can exactly meet the needs of individuals and teams for improving efficiency in creative assistance.

Finally, I have to mention the circle-breaking effect produced by Stable Diffusion’s open source. As a text-to-image generator, Stable Diffusion not only opens up the program, but also its already trained model, which means that it allows subsequent entrepreneurs to use open source frameworks to build a more open and vigorous content ecosystem. It provides the possibility for links to a wider range of C-side users.

2. Arousing the worry of "job grabbing" in the circle of painters

Although AIGC has many talents in the fields of design, painting, writing, programming, etc. There is room to show off one's skills, but doubts about itself have never stopped. Topics related to "AI painting" even appeared on the hot search lists of many platforms, becoming a veritable traffic password.

There is a huge polarization in the circle of painters. Some people are resistant or wary, while others accept it calmly and are even willing to take the initiative to contact them.

Opponents believe that the tireless AI "can produce a painting in 10 seconds, but it takes me several days." In the short term, its emergence will crowd out the career space of a large number of middle- and low-level painters. In the long run, it will lead to serious faults in the painting circle, and only the "big guys" will survive in the end.

What's more, it puts forward the point that "painting" is already an irreplaceable professional skill in the industry. If painters can be replaced, other professions will also be replaced. Will it be far?

Just like a netizen's comment: "I don't know where the future is. I believe that AI painting cannot completely replace humans, but it can replace me." This sense of crisis has caused many Human resonance.

Supporters believe that technological development is unstoppable, and AI painting can help painters produce content more efficiently, and sometimes even provide inspiration. Survival of the fittest is inevitable in development and will not be based on human emotions.

Of course, there are also those who hold a meaningless attitude towards this and claim that stones from other mountains can attack jade. For example, CG painter Wuhe Qilin said: "When it comes to painting, don't we always learn from whoever is good at it? If the AI ​​is good at painting, it's all over if we just learn the good parts from him."

Source: CG artist Wuhe Qilin comments @微博

Since the birth of artificial intelligence, one thing that cannot be avoided is : Concerns about “AI replacing people” need to be faced head-on. The development of AIGC seems to have intensified this "job" struggle, but if you return to AIGC itself amid the clamor, you will find that the significance of AIGC's existence is more to release producers from boring production work, or Help producers break through production bottlenecks rather than replace practitioners. As David Holz, the founder of the AI ​​painting tool Midjourney, once mentioned:

"I think artists are not just about the generator of images, art is often about stories and emotions, and AI has no stories and emotions. The narrative of the images created by AI comes from the people who use it. Shells all come from the sea, but the ocean is not the creator. Beautiful rocks all come from rivers, but rivers are not the creator. There is nothing created by this system Ability, but beauty can come from it."

The role of artificial intelligence in creative activities is only a tool and a carrier, not the creator itself. Holz compared artificial intelligence to water. “Water is dangerous, but you can also swim in it, build boats, and use dams to generate electricity.” The key is to learn how to use it and live with it. “It has no will, no malice. Yes, you can drown in it, but that doesn't mean we should ban water. It's really a good thing when you discover a new water source."

3. The shadow behind the scenery

Although many people still have doubts about the value of AIGC, some even criticize it for wandering in morality. and the edge of the law. But to a certain extent, black and red are also red. AIGC has entered the public eye and is gradually reshaping many industry scenarios, laying a foundation for awareness and attention. The market is also generally optimistic about the development potential of this track.

According to the qubit AIGC map, AIGC is now mainly used in text, images, videos, audios, games and virtual humans. Most of the start-ups involved are concentrated in rounds A to B. , including DeepMusic, Reflection Audio, Listening Intelligence, Caiyun Xiaomeng, rct.ai, Movie Spectrum Technology, Hyper Parameters, etc.

China’s existing AIGC industry chain division

@"AIGC/AI Generated Content Industry Outlook Report"

Domestic major companies Baidu, Tencent, Alibaba, ByteDance, NetEase, etc. have also invested in the AIGC field.

• Baidu launched the AI ​​painting platform "Wenxin Yige";
• Dreamwriter, a writing robot developed by Tencent, can write in 22 prescribed scenarios ;
• Lubanner, Alibaba’s AI online design platform, can help marketers produce banners;
• ByteDance’s Jianying and Kuaishou Cloud Cutting can both provide AI-generated videos;
• NetEase has launched the AI ​​arranging system "NetEase Tianyin";...

However, although the prospects are optimistic by most people, there are still many problems in the field of AIGC. Fewer problems.

First, key core technologies still need to be improved. Among the criticism surrounding AI painting and AI writing, many people point out that the content it generates is stacked and has uneven quality, and sometimes even creates a funny feeling similar to "stitch monsters". A recent study by Harvard University showed that DALL-E 2 did not understand the relationship within the text prompt at all, but only glued together several entities in the text, with an image accuracy rate of only 22%.

Second, the differentiation of needs has raised the development threshold. Taking AI writing as an example, AIGC can automatically generate text based on keywords and even achieve a certain level of fluency, but different scenarios have different writing needs. For example, the wording of instructions and advertising copy are very different, and the styles of news reviews and novels are also very different. Different needs naturally lead to different training data, which undoubtedly increases the cost and time of AIGC product development.

Third, large-scale commercialization still faces many challenges. Since the current artificial intelligence is still in the traditional weak artificial intelligence stage, it must rely heavily on humans to model actual problems. However, this method is quite unstable, and coupled with the performance limitations of hardware facilities, the generated results are not ideal. , and users’ lack of willingness to pay, which means it will take some time for AIGC to be commercialized on a large scale.

Previously, Kite, an AI programming tool that once had 500,000 monthly active users, announced its failure after persisting for 8 years, which is disappointing. Its founder, Adam Smith, attributed the failure to "the algorithm wasn't good enough" and "users didn't buy it." "The biggest problem is that state-of-the-art models don't understand the structure of code, such as non-native contexts." Second, it took them a long time to discover that "individual developers don't buy tools, only their managers might. But Just making developers write code 18% faster - this is not enough to impress them."

Fourth, the dispute over the copyright of AI works has not yet been concluded. Nowadays, many AIGC tools are free and open for use, but there are still many users who are worried about the legality of their copyrights. Since the entire industry is still in the exploratory stage, different countries have different policies, and different companies have very different approaches to dealing with related issues. Therefore, the whole world is still in a period of ignorance and chaos.

The core issues in the copyright dispute mainly focus on two aspects: First, can you obtain the copyright of what the AI ​​model creates? Second, can you use copyrighted data to train an AI model?

The lawsuit caused by GitHub Copilot not long ago is a typical case. In this class action lawsuit, programmers accused OpenAI of allegedly violating the open source license. OpenAI and Microsoft used the code they contributed to train the proprietary AI tool GitHub Copilot. Many people in the industry believe that the progress and results of this case may set a precedent for copyright ownership issues in the entire field of generative artificial intelligence.

4. The path forward with constant controversy

Once upon a time, people were not enthusiastic about the application of AI in creative activities, especially in the arts. fantasy. Because AI can only rely on imitation to acquire skills, and creation is undoubtedly a unique talent of human beings. However, the emergence of AIGC seems to have broken this self-confidence of mankind, so the argument that "art is dead" has emerged again. But just as the emergence of the camera did not bring the end of painting, so did the development of AIGC.

Although we still face the challenges of immature core technology, difficulty in achieving large-scale commercialization in the short term, incomplete relevant laws and regulations, and technical ethics on the road to development, this is basically a technology It is a common problem on the development path. Collisions are inevitable and it will take time to adjust. It cannot be denied that the development of AI technology will indeed improve human productivity. On the road to constant controversy, all we need may be a little patience and a little confidence.

Reference link:

https://www.8btc.com/article/6776768

https://baijiahao.baidu.com/ s?id=1748281279813972418

https://view.inews.qq.com/a/20221123A0403B00

https://www.huxiu.com/article/721434.html

The above is the detailed content of How has the controversial AIGC become a top player?. For more information, please follow other related articles on the PHP Chinese website!