On February 16, OpenAI launched the first Wensheng ** model SORA, which became a blockbuster hot spot in the technology circle overnight, and attracted Musk, Zhou Hongyi, Yang Likun and other bigwigs to discuss one after another.
It is reported that SORA can generate a smooth HD of up to one minute through a short prompt word description, or a static **. Judging from the **demo updated on the company's official website, these **present realistic scenes similar to movies, covering multiple characters, different types of actions and background details, etc.
"SORA is the foundation for models that can understand and simulate the real world, and we believe this capability will be an important milestone in the realization of artificial general intelligence (AGI)." "There is no doubt that SORA is the latest and most important AI product series released by OpenAI after GPT and DALL·E.
In this context, SORA-related concept stocks in the A** field have skyrocketed today, and 12 of them have achieved a 20cm daily limit. However, at present, SORA still has limitations, and due to factors such as high generation cost and higher computing power requirements, there may still be a long way to go before large-scale commercial application. What the market is worried about is that after this hype boom, will SORA concept stocks "return to the pre-liberation period overnight" like last year's ChatGPT concept stocks?
In the eyes of many industry insiders, SORA is undoubtedly a major breakthrough in the field of artificial intelligence, which not only demonstrates the advanced ability of AI to understand and create complex visual content, but also brings unprecedented challenges and opportunities to the content creation, entertainment and film and television production industries.
It is understood that SORA's advantages mainly lie in the accuracy and diversity of text input and conversion, strong language comprehension ability, generation in graphs, forward or backward expansion along the timeline, excellent device adaptability, and consistency and continuity of scenes and objects. At the same time, SORA is the basis for models that can understand and simulate the real world, and is an efficient means of simulation for the development of the physical and digital worlds and the metaverse, and is expected to play an important role in the fields of machine vision, autonomous driving, and robotics in the future.
After the release of the SORA model, the concept of AIGC in Hong Kong stocks took the lead higher. As of February 16**, a total of 44 of the 54 constituent stocks of the Hong Kong AIGC Index have achieved **. Among them, China Literature Group rose 1014%, Kingsoft rose 784%, Ali Pictures rose 698%, SenseTime rose 625%, Zhihu rose 587%, Bilibili rose 473%。
Today, A-shares opened the first trading day of the Year of the Dragon, and the concept of SORA was strong at the opening, and the related ** large-scale daily limit became the biggest bright spot in the market. Other related concept stocks such as multimodal model concept, ChatGPT, Web30, AIGC, virtual humans, etc. are also available.
Data**: wind
Specifically, some SORA concept stocks opened higher in early trading today, and then many stocks such as Danghong Technology, Insai Group, Wanxing Technology, and Silk Road Vision rose to the limit. In addition, according to Wind** data, today's new SORA concept index has risen by 1136%。Among the constituent stocks, a total of 30 shares rose by more than 10%, of which 12 shares achieved a 20cm limit, including Chinese**, Yidiantianxia, Guomai Culture, Haitian Ruisheng, Tianfu Communication, etc.
Data**: wind
According to the observation of the titanium industry, more than 10 A-share listed companies such as Danghong Technology, Insai Group, Yidiantianxia, and Wanxing Technology have disclosed relevant businesses in the field of generative models on the interactive platform in the past three months.
For example, Wondershare Technology said on the interactive platform on February 2 that the company's ** creative product Filmora can be used for all kinds of ** creation and editing, and the "Sky Canopy" large model is a multi-** large model with ** creative AI technology as the core, covering audio, image, ** and other multi-modal capabilities.
Danghong Technology said on the interactive platform on January 5 that the company has a self-developed AIGC toolset, released a solution to generate three-dimensional volume with static, and achieved up to 800 times visual lossless compression through point cloud model conversion and compression algorithms, so as to realize mutual switching between different modalities.
In addition to the field of ** generation, the field of computing power is also worth paying attention to. From a global perspective, the closed-loop loop of the computing power industry chain from upstream hardware, midstream server switches, and downstream applications is becoming clearer, and from the cloud side to the device side, from hardware to software, is showing a vibrant scene. It is recommended to pay attention to the core manufacturers of the global computing power industry chain first, and secondly, there is a lot of potential for end-side AI-related ***, and there are also good opportunities for localized computing power**.
At a time when the SORA concept was booming, some investors were worried that it was just a "flash in the pan". After all, SORA concept stocks are becoming extremely hot, but if this situation continues, and related companies are unable to come up with products or services for a long time, it is easy to "return to the pre-liberation period overnight".
Take the ChatGPT concept that exploded at the beginning of last year, there are as many as 100 domestic companies that will report that they are developing ChatGPT-related products and applications after the Spring Festival in 2023, or introducing ChatGPT technology into their own business, and there are a large number of Internet and technology giants, but most of the concept stocks can be described as ups and downs.
Taking Danghong Technology as an example, on January 31 last year, the company said on the interactive easy, "The company actively promotes AI technology to empower various industries, including AIGC, NLP and other ChatGPT-related technologies and products have also been gradually improved and applied in some businesses." Then the stock price went all the way from 2787 yuan rose to a high of 69 on March 2899 yuan. However, after the tuyere of ChatGPT, the share price of Danghong Technology entered a downward channel, and even came to a low of 16 yuan on February 6 this year, down 77 from last year's high14%。
Data**: wind
Coincidentally, today's daily limit of Yidiantianxia also borrowed the concept of ChatGPT at the beginning of last year, and said in February last year that the company has begun to combine the corpus output of ChatGPT with the advertising industry scene for secondary training for advertising business. But its share price reached 33 in early April last yearIt started a downward trend after the 60 high, and as of August last year, it has given up all of its gains.
Data**: wind
In addition, SORA still has weaknesses and may still be a long way from large-scale commercial application.
On the one hand, OpenAI said that SORA may have difficulty accurately simulating the physics of complex scenarios. may not be able to understand cause and effect; It can also confuse the spatial details of the cues; It can be difficult to accurately describe events that occur over time, such as following a specific camera trajectory, etc.
On the other hand, it is difficult to reduce the application cost of SORA. Some industry insiders pointed out that at the technical level, the SORA model adopts the architecture of diffusion model + transformer, and there is no substantial breakthrough in the basic technical principle compared with before. Therefore, SORA uses a huge amount of computing power to solve the problem of timing consistency between frames, that is, in SORA, not only the diffusion model is used to solve the generation of the same frame, but also the diffusion model is used to solve the timing generation between frames, which determines that its generation cost is relatively high in a short time.
In addition, in the absence of fundamental innovation in the underlying architecture, computing power plays an important role in it. In particular, compared with large language models, the inference of ** large models requires greater computing power support, but the current AI infrastructure construction such as computing power is not sufficient.
"The world needs more AI infrastructure, including wafer fabrication capacity, energy, data centers, and more, and people are not planning to build enough at the moment," Altman said. He added that "building large-scale AI infrastructure and a resilient ** chain is essential for economic competitiveness," and OpenAI will work to help.
Talking about the follow-up trend of SORA concept stocks, GF** said that for A-share related AIGC application companies, in the short term, from the perspective of investment, the theme significance of SORA is more significant. It should be noted that the commercialization process of SORA still needs to be observed. First, SORA's current limitations need to be addressed gradually; Second, topics such as copyright compliance and content ethics are still challenges that AIGC applications need to overcome, and Wensheng** is no exception. (This article was first published in Titanium**app, author|.)Zhai Biyue).