Behind the scenes of Deng Ziqi’s “Crossing the Red Sea” is Nvidia’s AIGC black technology!

In the past few days, the search engine war triggered by ChatGPT has caused sparks to fly between Google and Microsoft.

Even if we are watching from afar, we can deeply feel that this AIGC craze may forever change the way human society operates.

Yesterday, the second phase of New Wise Talk with the theme of "Generative AI Explosion" was officially launched.

New Wise Talk is hosted by Ms. Yang Jing, the founder of New Wisdom. The guest in this issue is He Zhan, head of Omniverse, NVIDIA China.

Generative AI Explosion

After the host and guests came out one after another, the new episode of New Wise Talk has finally started recording. The theme of this issue is "Generative AI Explosion".

In fact, during the preparation of the program, generative AI has already experienced After several generations of iteration and evolution, it can be said that it evolves faster than the program preparation speed.

In 2022, just as we humans are rolling around, AI is also quietly evolving. Therefore, 2022 has become the year of the explosion of generative AI.

There are two things that can fully prove the popularity of generative AI.

For example, the work "Killing That Shijiazhuang Man" by the well-known domestic rock band Wanmian Youth Hostel suddenly became popular on station B. The reason is that each of its lyrics has been matched with images by AI.

Another thing should be familiar to many people. An American game designer used AI to draw a painting, and then used the painting to participate in an art competition, and actually won first place. This is also the first time in history that AI has defeated humans in art.

## Similarly, the strong iteration of NVIDIA Omniverse also reflects this. Now, users can easily use digitalization to create digital twins and virtual digital people, which lays a solid foundation for the construction of the metaverse.

Deng Ziqi "Crossed the Red Sea", NVIDIA virtual stage blessing

Ms. Yang Jing made a very interesting statement: In the past six months, human beings have been rolling back and forth. , and AI has actually been involved on the stage.

For example, in Jiangsu Satellite TV’s New Year’s Eve concert in 2023, when Deng Ziqi sang “Gloria” from “Revelation”, huge waves suddenly appeared around her. This was powered by AIGC’s technology. generated with AR technology.

This visual presentation is something that hundreds of millions of viewers in China can see with their own eyes. It comes from the metaverse. Shocking.

And behind this stage, there are some black technologies from NVIDIA as support. He Zhan did the on-site decryption.

For example, this seawater special effect is an important application direction for content generation. This stage utilizes XR technology and combines it with the latest AIGC-generated technology to present us with a gorgeous visual feast. Behind these technologies, there are also advances in graphics.

The second black technology is NVIDIA’s optimization of AI deep learning acceleration algorithms. On the stage, it is the effect of large-scale training model iterative shaping.

For example, seawater requires a lot of simulations and more model training data sets. After it is finally presented on the stage, the audience can enjoy the beautiful singing while also being immersed in the scene.

Three nodes of AIGC

Based on the above background, Ms. Yang Jing asked He Zhan: Behind such rapid and strong iteration, in this wave of generative AI , what algorithm supports artificial intelligence generated content (AIGC)? What requirements does it have for algorithms and computing power so that young people can use it to surf and be cool?

He Zhan gave a wonderful answer to this question from several angles.

First of all, the concepts of AIGC and generative AI are not far away from us. There are several historical nodes that allow us to better grasp the development context of AIGC.

As early as 1957, there was the first string quartet composed by artificial intelligence - "Iliac Suite", which is the earliest artificial intelligence music.

When the time was pushed to 2007, New York University published a novel created by artificial intelligence. Although there are various logical errors and various vague plots in it, it is the first novel completely completed by AI.

The next time point is 2014. In this year, GAN (Generative Adversarial Network) appeared.

These three time points happen to be different iterative stages of generative AI or AIGC.

There is a very interesting thing in this. I wonder if readers have noticed it——

In the first stage of AI development, that is It took 50 years from the birth of the first AI-generated music "Iliac Suite" to the first AI-generated novel; but only 7 years passed from the first AI-generated novel to the emergence of the GAN network .

In recent years, especially in the past six months, generative AI has sprung up like mushrooms after a rain. There are text-to-image DALL-E2, Midjourney, Stable Diffusion, etc. These models are iterated in a week or two, which is very fast.

The rapid iteration of deep learning technology, including the generation of GAN just mentioned in 2014, has greatly accelerated the development of generative AIGC technology.

How should young people use AIGC’s technology? In He Zhan's view, the future is endlessly imaginative. We can all see that there are more and more jobs covering creative content.

Including the AI ​​mentioned earlier for writing articles, making music, post-production, etc., these various applications may provide young people with a very large space to explore and explore.

If young people embrace these changes and continue to iterate their abilities, they will have unlimited potential in the future.

Trillions of Dollars Blue Ocean

2022 is the year when the metaverse explodes, but there are still Before the Metaverse could catch up, the new concept of AIGC became popular on the Internet with lightning speed.

Partners of Sequoia Capital even co-wrote an article with GPT-3, predicting that AIGC will form a new track worth trillions of dollars.

Ms. Yang Jing asked: Why will AIGC take advantage of the east wind to explode in 2022? What is the technical logic and industrial logic behind it? And what are the similarities and differences between the Metaverse and AIGC?

He Zhan explained that when talking about GPT-3, we must mention the parameters behind it-175 billion.

When GPT-3 was first released, many researchers and developers were stunned. Just a week ago, there was a report about the warm-up of GPT-4, which mentioned that the parameters of GPT-4 could reach 100 trillion.

For a model that has entered the 100 billion parameter level, what will the content created by it look like in the future? This is something everyone can look forward to.

GPT-3 and GPT-4 play the role of content production in the Metaverse.

If you want Metaverse applications to iteratively develop rapidly, you must involve as many people as possible.

For example, if you want everyone to participate in the Omniverse platform, you must lower the technical threshold so that everyone can use it to create more works.

At this time, you need tools that can be quickly generated, fast, good, and low-cost.

So where does productivity come from? It is a tool from the generative AIGC.

Now, many companies, including Nvidia, are making conferencing systems. There is Maxine algorithm in video conferencing. If you want to drink water while driving, the algorithm can correct your face into your new face when you go to drink water, making others think that you have been staring at the screen. .

Um? Wait, isn’t this fishing?

Yes, in fact, this is fishing.

A few weeks ago, Nvidia had an interesting user exchange.

Some users said that they want such a live broadcast room, which is a study room. There should be some photo frames or artwork in the book. There should be blue sky, white clouds, etc. in the photo frame.

In fact, this kind of requirement is not as demanding as the studio stage, and it is completely achievable now.

For example, for the wallpaper in the study, you can use the tool to input whatever style and tone you want, and it will produce real-time effects.

Nowadays, these technological advancements have completely stimulated personal creativity.

Imagine, in the past you had to draw an oil painting, but now you can directly let AI generate it quickly. This is really dreamy.

2023: The first year of AIGC

Ms. Yang Jing said that this idea is really exciting, because now behind the video accounts on Weibo and WeChat, They all have tens of millions or even hundreds of millions of users. If AI can be used to generate special effects or videos, it will undoubtedly stimulate many people’s desire to create.

So, can this wish come true in 2023?

He Zhan cited a report from a well-known research organization. According to this report, the current proportion of generative AI in the entire artificial intelligence generation is less than 1%. Therefore, if it can reach 10% in 2025, it will be an amazing result.

In the field of bioscience and medical care, by 2050, the proportion of drugs and materials generated by AI may reach 30%.

So, among the various generative AI technologies, which ones can be the killer? When will AIGC really go to the public and become a super APP, and which companies will win the golden opportunity?

He Zhan believes that the most critical thing is to identify which killer applications are needed for these applications.

For example, what should you do if you need to design an electric kettle but want to be lazy?

Now, there are actually many 3D model generation tools from major manufacturers that can customize this kind of design.

For example, Google's DreamFusion and NVIDIA's Magic 3D can generate the effect you want by just entering a piece of text.

In summary, AIGC can break out at any time as long as it can meet the needs of designers or engineers.

One-click generation of cool dramas

Obviously, generative AI is more closely integrated with content. AI can be used in e-commerce, media, film and television and other industries. Assists in video script creation, game scene generation, digital human-assisted delivery, XR product display, etc.

Now, a few words can generate a script, or even a short video or short movie. And if AIGC is implemented on a large scale in the future, which link in the industry chain will have the greatest impact?

He Zhan replied that the Shuang drama that has become very popular on video accounts recently was filmed using Shuangwen. But the productivity in this process is actually lacking.

You must know that there are about 100,000 production staff behind Shuangju in China. These 100,000 people are at the end of the entire production chain, and their works have been handed over to many hands. Only then does it truly become a work on stage or screen.

These production staff worked very hard but did not get much benefit. However, if Shuangju can be quickly produced through AIGC, a group of people will come in. When the technical threshold is lowered, productivity will rise, and there will naturally be a closed loop.

Moreover, not only screenwriters, but also in fields such as drug research and development, education, etc., will have such a closed loop of production, release, and economic benefits.

However, Ms. Yang Jing raised a very critical question: Since cool articles and cool dramas can be generated with one click, will young people become too dependent on such tools and lose their money? What about imagination?

He Zhan said that it is certain that AI tools will not make people lose their imagination.

For example, Nvidia held a design week event in Hangzhou last year, and displayed a craft called "Magic Pen Ma Liang" - Nvidia Canvas. Outline a curve on the left side, and on the right side, you can use your input to let the AI ​​make up a picture.

Simply draw a few strokes and a real photo-like work will be generated on the right.

What moved He Zhan very much was that many of the children present were much more serious than the adults and were very attentive. And many adults are joking or have a shy attitude.

Therefore, the paintings they drew were more imaginative than those exhibited by adults.

The same AI tool produces completely different works. This tells us that if the input imagination is different, the effect achieved by the work is completely different. Therefore, even tools iterated by technology are still inseparable from human imagination and concentration.

Therefore, generative AI can stimulate the imagination of young people, free their thinking from being constrained, and let their imagination take wings. No matter how amazing the tools that appear in the future are, the final input still depends on ourselves.

Domestic players are vying for the throne

Speaking of this, we have to mention the recent vying for the throne by major manufacturers such as Google, Microsoft, and Meta. New unicorns like OpenAI have also recently attracted attention with ChatGPT and received more than US$10 billion in investment from Microsoft.

At the same time, domestic major manufacturers such as BAT are also rushing ahead in the AIGC field.

Which players will be the frontrunners? What is the biggest highlight of technological development in 2023?

He Zhan believes that major domestic Internet companies will definitely have models similar to ChatGPT.

For example, Alibaba’s online shopping, Tencent’s social networking, etc., there are many Lenovo applications, and big manufacturers will definitely invest heavily.

Planet-level computing power can generate a dynamic "Book of Life"

At the end of the program, Ms. Yang Jing talked about a magical dream she had recently. In the dream, her classmates gave her a photo album, which vividly presented the most memorable scenes in her life with numbers and images, like a living book of life.

From this, Yang Jing had a wonderful idea: Can AIGC technology be used to automatically generate a virtual photo album from a person's images for a year or a lifetime? If we want to review our Weibo or Moments, it will be difficult to find these memories day by day. However, if we use AI to review these scenes and automatically generate a book of life, it will be much easier.

He Zhan said that the problem is not big, just like mobile phones sometimes push us a moment, and the same logic applies.

We can use our data over the past few years as input, and then input and generate it. And you can request to be more tender or cheerful, and the generated photo album will also have corresponding emotional expressions.

## Ms. Yang Jing talked about a puppy she once raised named Xiaodoudou, who died in 2020 . She once saw an advertisement where if you input a photo of a dog into the product, it would generate a photo album of the dog's life. But it would be troublesome to specifically look for photos of dogs, so if AI can automatically search for them from the gallery, it will definitely meet the needs of many people.

He Zhan was very moved after hearing this. He also believed that as tools develop faster and faster, they will enter more and more into ordinary people’s lives, and there will definitely be new ones. As industries emerge, for example, some people will use these tools to create new apps.

Yang Jing said that our partners, pets, and relatives are all the weaknesses and the most tender parts of our lives. They are the greatest assets of human nature.

In addition to cool articles and cool dramas, the future metaverse will also have large-scale virtual cities and many virtual stars with different personalities, who are not even human beings.

These digital clones of virtual and real stars can generate new digital photo albums and virtual movies, thus creating countless life books. In the future, AIGC will have the ability to generate a new digital planet. , Digital Earth, or even an all-encompassing Digital Metaverse?

He Zhan believes that everything is possible. Imagine that all these technologies ultimately promote productivity. Now, everyone’s needs already exist, such as generating a photo album, so that People look back on those tender moments. What users need is fast, good and cheap. For example, a photo album can be generated for more than ten yuan.

At the end of the program, Ms. Yang Jing concluded: Huang Renxun’s mantra is “saving money”, so a fast, good and powerful generative AI will definitely be able to Subvert the future of mankind.

So, in 2023, we are grateful to generative AI for taking us across the Red Sea and taking us to review the warm memories in our lives, so we are even more looking forward to the 2023 Generative AI Conference The explosion will bring about a colorful world, as well as a metaverse and a new universe full of infinite possibilities and planet-level computing power.

The above is the detailed content of Behind the scenes of Deng Ziqi’s “Crossing the Red Sea” is Nvidia’s AIGC black technology!. For more information, please follow other related articles on the PHP Chinese website!