Recently, openai’s official website released the Wensheng video model sora, which can create realistic and imaginative scenes based on text instructions and generate videos up to 60 seconds long, triggering heated discussions in the industry. Since OpenAI's current information release on the current status of the model is not complete, discussions in the industry still focus on its innovation in technical details, impact and influence on the film and television industry and other related fields. Nandu reporters noticed that there are many "cross-overs" in the video case scenes released so far. As the most advanced Vincent video model currently, the
sora model has any known technical flaws? Will the technological innovation of Vincent Video Model cause disruptive changes to the development and governance of the existing artificial intelligence (AI) industry? In this regard, domestic and foreign professionals still believe that the development of Vincent video models is still in the "gpt3" era and "is only the first step." An expert in the field of intellectual property and technological security analyzed to Nandu reporter that at present, the technology and governance logic of the Vincentian video model and the Vincentian graph model are still similar, and it may be possible to look forward to the subsequent development of the New York Times' lawsuit against Openai for infringement.
sora model release case video has technical flaws
Nandu previously reported that on February 15, local time, openai’s official website released the Wensheng video model sora, which can create realistic and imaginative scenes based on text instructions and generate 60-second videos, triggering Hotly discussed in the industry.
video demo screenshot.
In Sora's technical report, OpenAI calls it "a video generative model as a world simulator." OpenAI said they explored methods for large-scale training of generative models on video data. Its largest model currently, sora, is capable of generating 60 seconds of high-definition video. According to foreign media reports, compared with other industry video models, 60 seconds is the current highest record. The previous record of the longest AI-generated video in 2023 was 16 seconds.
has 48 video demos on the sora related introduction page, with corresponding prompt words. Openai stated that all videos on this page were directly generated by sora without modification.
openai also pointed out that the current model still has weaknesses. It may have difficulty accurately simulating physical phenomena in complex scenes, it may fail to understand specific cause-and-effect relationships, and it may confuse spatial details in cues. The model may also have difficulties in accurately describing events that occur over time.
Nandu reporter noticed that in one of the widely circulated Tokyo street videos, the subject of the video - a woman - suddenly exchanged positions with her left and right legs while walking. In the background, some pedestrians had other people or objects in front of them. Disappears after passing. In addition, other videos also appeared in other videos, such as an old lady blowing out a candle but the candle did not go out, and a man running in the opposite direction on a treadmill.
Currently, the sora model is not available to the public, and OpenAI is working with red teamers. Red team members refer to people employed by organizations or institutions in the field of network security to simulate attackers for penetration testing and security assessment. They will conduct adversarial testing of the Sora model. Visual artists, designers, and filmmakers alike are also allowed to access Sora and provide feedback.
According to foreign media reports, artificial intelligence videos have already caused a sensation in Hollywood. Some tech experts theorize that artificial intelligence could allow a person to create a movie with the visual complexity of a Marvel blockbuster. Many film workers have also publicly stated that they feel confused about their future. American entrepreneur Elon Musk later responded to tweets about the release of the Sora model on social platforms: gg humans (human beings are finished).
may follow the governance logic of the Vincentian graph model
Nandu previously reported that on December 27, 2023 local time, the New York Times filed a lawsuit against openai and Microsoft, accusing the two companies of using the media for hundreds of millions of dollars without authorization. Thousands of articles train chatgpt and other ai chatbots.
The New York Times stated that in some cases, OpenAI and Microsoft copied the content they published verbatim and provided it to users seeking answers from AI chatbots. These new products have created potential competitors for news publishers, and This type of use of copyrighted media works is illegal.
The New York Times sued Microsoft and OpenAI for infringement.
"This case may bring major changes to the existing copyright law in the United States," Hao Min, director of the Intellectual Property and Technology Security Research Center of the School of International Relations, analyzed to Nandu reporters. At the same time, as a country with case law, the United States has The verdict of the case will also bring "logical inheritance" to the governance of subsequent products of similar nature in the AI industry, such as Vincent's video model Sora.
Hao Min said that in the future, countries will form a "balance of interests" when it comes to intellectual property laws related to the AI industry. “Whether it’s the New York Times’s protection of its existing written works and its requirement for OpenAI to pay for the use of its written materials, or OpenAI’s consideration of limiting the training materials for AI models as a leading company, which is detrimental to itself and other small and medium-sized related companies. There is a certain truth to the negative impact of enterprises. This is a dispute between industries and between interest groups, and it will ultimately be the result of weighing the pros and cons."
"The current core view in the legal profession is the 'tool theory', and AI is still a 'tool' to assist people in their actions," Hao Min pointed out, "The subject of intellectual property rights must be natural persons, not the machine itself."
And sora The impact and influence of the model on the existing film and television industry may also be predicted from the perspective of intellectual property rights. Many industry experts at home and abroad mentioned that if the image data collection of film and television industry practitioners by Wensheng video models such as Sora is one-time and permanent, it may cause the extras profession to become a one-time data collection work.
"In fact, the impact is similar to work such as celebrity endorsements. The collection of celebrity portrait data may become a one-time job. This comes back to the use of data calculations. For the enjoyment of intellectual property rights Is the calculation of the data considered fair use, and do I need to obtain consent and pay uniform remuneration for use? Many rights issues will also be discussed around the use of portrait rights." Hao Min told Nandu reporters.
The release of the sora model, which has caused heated discussions recently, is still the beginning. Many industry experts at home and abroad pointed out that the development of Wensheng video model is still in the "gpt3" era, and it will continue to follow up on governance and supervision while constantly making up for technical shortcomings.
Written by: Nandu Trainee Reporter Xiao Yue Intern Chen Nana
