Source: Gelonghui

Sometimes the feeling of powerlessness in life is that you know that the world is undergoing great disruption, but you are unable to participate in this transformation. You can only watch other people's carnival. Even if you know the near future, your life will definitely be held hostage to move forward as a result.

Open AI is such a great company that changed the lives of all people. After the text model ChatGPT and the image model Dall-E slaughtered the four sides, OpenAI continued to lay out the big killer, Sora. This time, they chose to disrupt the video field.

1. What is Sora?

On February 15, Open AI released the first Wensheng video model, Sora, which can generate one-minute high-fidelity videos, stirring up thousands of waves with one stone.

Open AI claims that Sora is a “world simulator.”

People exclaimed, “The real world doesn't exist anymore.”

Musk put it bluntly: Humans accept gambling and lose.

Former Alibaba VP Jia Yangqing commented on Sora: It's really cool.

Jim Fan, a senior research scientist at Nvidia and acting head of artificial intelligence, believes Sora represents the GPT-3 moment of text-generated video.

What exactly is Sora? What are the advantages compared to similar products such as Runway and Pika?

Currently, there are two main types of mainstream technology routes for live video models on the market: one is based on the Transformer model, that is, generated from text and images, and the other is based on a diffusion model (Diffusion model), such as Runway.

Sora is a Diffusion Transformer model that combines the two. By combining the diffusion model (DALL-E3) and converter architecture (ChatGPT), Sora doesn't need to predict the next text in the sequence, but rather the next “patch” in the prediction sequence.

This means that Sora is trained based on a “patch” rather than the entire video. It's a bit like ChatGPT processing videos with tokens. Therefore, Sora can efficiently process more data, and the output quality is higher.

In fact, the most impressive feature of the demo video released by Sora was that it realistically simulated the physical world, and the video effects applied to similar products such as Runway and Pika on the market.

The team behind Sora once again recently announced a new video production. This time, even multi-angle cameras in the same scene have appeared.

2. Who is Sora's victim?

It's only been a year since ChatGPT started the generative AI era, and we're still learning how to use indicative sentences to make better use of ChatGPT. When people doubt that AI is getting lazy, Sora's appearance can be described as a complete surprise. Now everyone is beginning to doubt the boundaries between the real world and the virtual world.

The evolution of AI far exceeds expectations. Needless to say, AIGC has the potential for disruptive innovation. The existing industrial pattern, such as short videos, advertising, games, film and television, etc., will definitely be reshaped. Who bears the brunt of this?

Judging from the performance of the secondary market, it is a tool company$Adobe (ADBE.US)$The day after Sora's announcement, the stock price plummeted by more than 7%.

Another US stock$Shutterstock (SSTK.US)$It fell more than 5% last Friday, and the market value evaporated by more than $70 million overnight. According to public information, the company sells photos and videos worth around $1 billion a year.

Also, Sora's appearance directly made netizens all exclaim: “Will online novelists be able to directly produce movies for us to watch in the future?” Another netizen directly hit the nail on the head: “Do you think there will be any online novelists if you continue to develop? Is ChatGPT not working?”

Also, AI is all this raging. What do you want, directly generate it for you with one click. Does a company that sells images have any value?

Just how much influence did Sora have?

3. How much influence does Sora have?

360 Chairman Zhou Hongli commented: Today, Sora may bring huge disruptions to the advertising industry, movie trailers, and short video industries, but it won't necessarily defeat TikTok so quickly; it is more likely to become TikTok's creative tool. Furthermore, he believes that the artificial intelligence gap between China and the US is widening.

Jia Yangqing, a former VP of Alibaba and currently starting a business at the AI Infra Circuit, expressed opinions in her circle of friends that, first, “companies targeting OpenAI have a wave of opportunities to be acquired by other major companies FOMO (afraid of missing out)”; secondly, “the long-term closed source oligopoly, open source will take some time to catch up”; finally, small algorithm manufacturers “will either be comparable to OpenAI in terms of algorithms, or they will further cultivate applications in vertical fields, or not, open source” and predicted that “demand for infra will continue to soar.”

From an investment perspective, if you target the wave of AI speculation on A-shares at the beginning of last year, computing power and gaming are the most prominent.

What is particularly noteworthy is that prior to Sora's announcement, there was news that OpenAI CEO Sam Altman would raise 7 trillion US dollars to completely reshape the global semiconductor industry.

What is the concept of $7 trillion? This is basically the entirety of the global semiconductor industry. With this money, Altman can directly package and take away all companies such as Nvidia, AMD, TSMC, Broadcom, ASML, Meta, Samsung, Intel, and Qualcomm.

Is this Ultraman coming to save the Earth?

Altman said in a post on X that OpenAI believes “the world needs more AI infrastructure, including wafer manufacturing capacity, energy, data centers, etc., and people are not currently planning to build enough.” He added that “building large-scale AI infrastructure and a resilient supply chain is critical to economic competitiveness,” and OpenAI will work to help.

In response, Nvidia founder Hwang In-hoon responded: I am optimistic that the global AI data center will double and grow to a scale of 2 trillion US dollars in the next 4-5 years. In this period, more efficient and lower cost chips will continue to appear, and large-scale investments will become less necessary.

What is undeniable is that large model products have risen from text and images to videos. The amount of AI training continues to increase, and demand for computing power is also growing exponentially. Companies that “sell shovels” are still the number one beneficiaries.

Editor/Corrine