Our reporter Zhao Yuhan and Yuan Lu

  At a similar time, it was even more shocking than last time. On February 16, OpenAI, an artificial intelligence technology company that once shocked the world with its large model ChatGPT, once again launched a new "king bomb" - Vincent video large model Sora. Just enter a text command to generate a 60-second video. The picture is exquisite and delicate.

  "AI (artificial intelligence) is growing faster than imagined!" Some people are excited about this, while others express concerns: "Are humans really ready?" But more and more voices believe that artificial intelligence will lead the fourth Second industrial revolution.

The first large video model was born

  Sora is not the first Vincent video model. Previously, technology giant Google and startup Runway have already laid out this track. Why does Sora amaze the world? Compared with previous applications, it greatly extends the time to generate a video to 60 seconds, and has also significantly improved the details, smoothness, and multi-angle presentation of three-dimensional space.

  Zhou Hongyi, the founder of 360 Group, talked about several shots that impressed him: a puppy playing in the snow, splashing snow particles, and its furry hair was clearly rooted and floating in the wind; two sailboats were "entangled" in a coffee cup. "The bucket" stirs up surging water ripples; street scenes from different angles are smoothly presented through the movement of the mirror.

  Beyond the video field, deeper changes may be happening quietly.

  Zhou Hongyi said that Sora's subversiveness is not limited to a text-generating video tool, but another breakthrough in general artificial intelligence's understanding of the world. A year ago, ChatGPT marked a breakthrough point in artificial intelligence’s understanding of human language and knowledge. In addition, there are many laws in this world, such as the snow that a puppy sticks on its nose, how the perspective changes when a drone flies in the air, etc. The video currently generated by Sora shows its influence on this world. There is a more complete understanding of the laws. This means that the era of general artificial intelligence may be approaching at an accelerated pace.

  Xie Xuefang, a distinguished professor at the School of Humanities at Tongji University, also believes that the emergence of Sora means that AI’s ability to understand the real world has further improved, but the breakthrough development of such large models will also be restricted by computing power, algorithms, and ethical regulations in the future, and it should be maintained Positive and cautious attitude.

The video industry landscape may be reshaped

  The emergence of Sora has made practitioners in the video field feel a crisis. "It's so shocking, I feel like I'm going to lose my job." Hu Juan, founder of a cultural communication company in Beijing, said frankly that Sora made her sense of crisis sharply increase. "This can be used for lens prototypes and some complex special effects production. In the past, the construction period was at least It took several days, but the AI ​​reduced it to minutes."

  Photographer Li Haonan believes that Sora will have a greater impact on film and television-related industries and can significantly reduce production costs, especially videos involving multiple angles and multiple cameras. If AI can partially replace manual shooting and editing, the cost may be only ten percent of the original cost. one part.

  Sora's technological breakthrough has also made the prospects for generative AI startups and investment institutions large and small very challenging. Luo Chao, operating partner of a venture capital firm focusing on global early- to mid-stage companies, said that Sora’s leadership has exposed star companies such as Runway and Pika Labs that generate large-scale video models. Compared with these self-developed large-scale model companies, pure model companies are facing a strong impact. The future of application startups is even more uncertain.

  According to Lei Tao, CEO of Xinyi Technology, the birth of Sora will undoubtedly change the market structure of AI video generation. "As far as the country is concerned, it depends on which companies can combine and utilize new technologies to improve products that suit the national conditions."

AI makes "seeing not necessarily believing"

  The generated video shown by Sora was also found to have many "crossing" scenes. For example, in a video of an old woman blowing out a candle, the candle did not go out after the old woman blew out the air; in another video of a glass falling from the sky, water was already flowing out of the table as soon as the glass was lifted into the air.

  This is also the shortcoming of the current Vincent video model - it does not understand the physical world and causal relationships. According to Zhu Wei, deputy director of the Communication Law Research Center of China University of Political Science and Law, the video generated by the large model represents AI's understanding of the laws of the world, but this is not necessarily correct, so "seeing is not necessarily believing." .

  A broader concern is that videos are more trustworthy than text and pictures. Once the threshold for generation is significantly lowered, fake videos whose authenticity is difficult to distinguish may be used in illegal fields such as spreading rumors and fraud. In this regard, Zhu Wei believes that the content generated by AI should be labeled with reference to previous relevant regulations on generative AI.

  In addition, after the creator uses a piece of text to generate a video, issues such as the copyright ownership of the video also need to be reorganized. Currently, there is a debate over the copyright ownership of AI-generated text or images. At the end of last year, the Beijing Internet Court issued a first-instance judgment on a copyright infringement dispute case involving "artificial intelligence generated pictures". This was also my country's first copyright case involving "AI generated pictures". The judge believes that when people use generative artificial intelligence to generate pictures, they are still essentially using tools to create. If it can reflect people's original intellectual investment, it should be recognized as a work and protected by copyright law. (Beijing Daily)