Screenshot of OpenAI official website
Input only a few words to generate a 60 second video with burst effect! The video generation model Sora launched by the Open Artificial Intelligence Research Center (OpenAI) has shocked the world recently.
OpenAI just launched ChatGPT more than a year ago, and it threw a heavy bomb again. What is the amazing place of "Wensheng Video" AI model Sora? What is the impact on the industry? What are the hidden dangers? Xinhua News Agency reporter will solve the problem for you——
What is Sora capable of
Several short video "blockbusters" generated by Sora have been screened on the whole network, with realistic silky scenes and rich details.
This is the first time that OpenAI has entered the field of AI video generation. According to the company, Sora uses the Transformer architecture to create realistic and imaginative scenes according to text instructions, and generate high-definition videos of various styles, different frames, and up to one minute in length.
In other words, if Sora is given some hints, it can produce a high-quality short video. Let's experience Sora's creative strength.
The Sora generated video image released on OpenAI official website shows that an SUV is driving on a winding mountain road.
Sora can also generate video based on still images, expand existing video or fill in missing frames.
OpenAI said that Sora can deeply understand the language, not only understanding user text prompts, but also understanding the existence of the said things in the physical world. "We are teaching AI to understand and simulate the physical world in motion. The goal is to train models to help people solve problems that need to interact with the real world."
However, is the video works published on OpenAI's official website the average or the highest level created by Sora?
OpenAI acknowledges that at present, Sora generated videos may contain illogical images, confuse left and right spatial details, and it is difficult to accurately simulate the physical principles and causal instances of complex scenes. For example, a person may have bitten a biscuit, but there is no bite mark on the biscuit. However, with the enhancement of computing power and the improvement of models, people may obtain more complete and advanced video generation functions in the short term.
Whose jobs may be destroyed
The Sora launched by OpenAI this time is more like a preview version, and it is difficult for the public to fully understand the advantages and disadvantages of this model. OpenAI said that at present, only some designers and film Specific groups such as producers provide access to Sora to obtain feedback on improving the model. Not only did the company not release basic details such as data for training Sora model, but also did not determine when to release it to the public.
Sora generates video screenshots
Some analysts believe that Sora once again highlights the profound impact of AI technology progress on real life and traditional industries. The huge development prospect of AI in the field of video generation opens the door to shaping a new format of the film and television industry, and at the same time, it is feared that the existing film and television industry will be subverted.
The second day after Sora launched, the stock price of Adobe, which is mainly engaged in image processing and video production software, fell by more than 7%.
Last year, Hollywood suffered the first industry wide strike of writers and actors in 63 years, because some jobs in this industry may be replaced by artificial intelligence. Sora's birth makes this threat more imminent and real.
Make forgery more realistic
OpenAI wrote in the title of Sora technical report that the video generation model is a "world simulator".
If the world can be simulated, where is the boundary between true and false? Many insiders worry that Sora will contribute to the "deep forge" technology. "When Journalism , images, audio, video - when anything can be forged, nothing is real in that world. "
The picture shows that on November 2, 2023, a participant passed the promotional panel of the first AI Security Summit in Bradley Park, UK. Photographed by Xinhua News Agency reporter Li Ying
In response to the concern of fraud, OpenAI said that when it really launches products to the public, it will ensure that the generated videos contain source metadata, and launch tools to detect the authenticity of videos. OpenAI also promises to take security measures before using Sora in products, including adversarial testing of models by "experts in the fields of error information, hate content and prejudice" to assess hazards or risks; Check and reject text input prompts including extreme violence, sexual content, hate images, others' IP addresses, etc.
However, OpenAI acknowledges that even after extensive research and testing, "we cannot predict all the beneficial ways people use our technology and all the ways people abuse our technology".
Can AI Crazy Racing Be Handled
Disruptive innovations continue to emerge in the field of science and technology. How to achieve the balance between embracing technological progress and ensuring social security has attracted more and more attention from all walks of life.
OpenAI said that it would work with policymakers, educators and artists around the world to understand their concerns, identify positive use cases of Sora, and believe that learning from the use of the real world is a key component of creating and releasing increasingly secure AI systems.
On July 7, 2023, at the "Global Summit of Artificial Intelligence for Humanity" in Geneva, Switzerland, a participant interacted with Sophia, a simulated robot. Photographed by Lian Yi, a reporter from Xinhua News Agency
Industry insiders pointed out that in the absence of the current governance framework and control measures, enterprises alone may not be able to provide the AI security and trust required by society.
(Reporter: Wu Xiaoling; Editor: Chen Dan, Diao Ze, Wang Kewen, Hu Dandan, Li Xuemei, Chen Yong, Hu Dandan, Sun Jing, Xu Chao, Ma Zhen, Zhang Jiawei, Lin Xiaochun, Xu Xiaolei, Du Sege, Liang Shanggang, Wang Fengfeng)
(Source: Xinhuanet)