On February 16th, OpenAI announced the launch of its new artificial intelligence model, "Sora." It is understood that Sora can directly output videos up to 60 seconds long via text commands, featuring highly detailed backgrounds, complex multi-angle shots, and multiple emotionally expressive characters.
Following text and images, this marks OpenAI's expansion of its advanced AI technology into the video field. Sora is undoubtedly a major breakthrough in artificial intelligence, demonstrating not only AI's advanced capabilities in understanding and creating complex visual content, but also bringing unprecedented challenges and opportunities to the content creation, entertainment, and film production industries.
Watch all the amazing videos released by Sora in one go!
01
Thoughts on the origins of Sora
We know that humans identify objects, thereby understanding and changing the world, by acquiring and recognizing various information from nature and society. Information broadly refers to all content disseminated in human society, including audio messages, news reports, and the objects transmitted and processed by communication systems. This information exists in various forms, such as text, symbols, tables, sound, graphics, images, video, and animation.
However, the emergence of Sora now means that the understanding, simulation, and generation of the real world are no longer limited to humans. It encompasses forms such as text, graphics, images, voice, and video—capabilities that were previously only possible for humans. Therefore, Sora's advent heralds an unimaginable transformation taking place in the world.
02
The core technology of large models: machine learning
The core technologies of large-scale models are natural language processing (NLP) and deep learning. Specifically, it is based on the Transformer architecture and uses large-scale unsupervised learning methods, such as autoregressive language modeling and masked language modeling, to train a massive neural network model.
Machine learning sounds sophisticated, and it is indeed very complex. To complicate simple problems, machine learning simulates or implements human learning behavior. If we understand machine learning from the perspective of human learning behavior, it may be easier to grasp.
In the process of understanding the world and learning, people usually start by learning abstract concepts to understand a thing or principle. This concept is actually equivalent to feature engineering in machine learning. For example, when we learn about the geometric symbol for a triangle, we first recognize it from the symbol, and then learn the definition, properties, and calculation formula of a triangle. The definition, properties, and calculation formula of a triangle correspond to feature engineering in machine learning.
Teachers typically assign homework after class to reinforce students' understanding of the concepts, definitions, properties, and calculations of triangles. The process of students completing these assignments serves as training data for machine learning. Training data is data used to help machine learn knowledge and build models that contain that knowledge.
To improve their learning outcomes and grades, students often attend extracurricular tutoring classes and engage in various practice problems and mock exams. This process provides machine learning with more and more effective training samples. Each semester, schools typically schedule weekly, monthly, midterm, and final exams to assess students' learning progress and serve as the basis for evaluation. The ultimate goal is to get into good high schools and universities.
Students taking various exams are like test samples in machine learning. Test samples are the objects that machine learning models serve; making correct predictions and providing feedback on test data is the ultimate goal of all machine learning activities.
We see that all the technologies and methods in the computer world, including machine learning, originate from our daily lives and practices, and are easily understood by the general public.
Many times, our parents and grandparents may not have been to school or even be literate, but they can understand and even naturally follow some of the ancient knowledge and wisdom of the Chinese nation (such as idioms and proverbs). This knowledge and wisdom is actually our genes, which are passed down from generation to generation and are not limited by written language.
This genetic inheritance can be imagined as a piece of "code." If this code was designed by a higher-dimensional world, then humans could be considered "artificial intelligence" of that higher-dimensional world. Now, based on their own genes and daily practices, humans have designed the N+1th generation of artificial intelligence, which in turn perpetuates the human gene chain in a different way.
03
What changes and disruptive impacts will Sora bring?
The introduction of the Sora model will undoubtedly have a significant impact on future video production methods and bring new opportunities to various industries. However, the impact of Sora, in terms of both breadth and depth, will far exceed these.
1. Impact and disruption on industries and professions
Based on OpenAI's official articles and shared demonstration videos, Sora can be described as a "game-changer" AI tool. Just as OpenAI integrated its image generation model DALL-E into its ChatGPT product, if OpenAI also integrates its video generation model Sora into ChatGPT for all users, it should gradually bring about changes to video-related industries, such as film and television production, short video creation, online education, advertising and media, game development, live streaming, and so on.
2. The impact of virtual reality integration
Sora mode can generate various scenes and scenarios in a short period of time based on user preferences and actions, enhancing the user's interactive experience. With the advent of Sora, when people are immersed in a virtual metaverse, such as with AR or VR, when the experience is taken to its extreme, they may not be able to distinguish between reality and virtuality. Or they may know but not want to distinguish, just like people who are in a beautiful dream and don't want to wake up, because life in the virtual world may be better than in the real world.
3. The integration of data and reality drives the elevation of spiritual realm.
With the development of digital technology, all processes of human social activities can potentially be recorded and formed into data. The emergence of Sora means that in the future, based on the labels and behaviors of value entities (such as organizations, enterprises, individuals, and groups), it will be possible to easily generate and reconstruct the original models and trajectories of the entire real world.
In other words, it might be possible to generate relevant videos (not just still images) directly, without needing to review camera footage or have forensic experts create a profile of a suspect based on witness descriptions. Therefore, in the future, the amount of wealth, possessions, and real estate you own will be transparent on the platform, and every penny you spend will also be transparent.
The potential consequences of this are that any expense will become public knowledge. If the money is ill-gotten, there might not even be an opportunity to spend it, and illegal or criminal activities will be even harder to hide. Therefore, the development of digital technology will drive people to improve their spiritual and moral cultivation, potentially leading them to shift their focus from material pursuits to spiritual exploration and discovery.
Sora is just a small test; it demonstrates more than just video production capabilities. It shows that once large-scale models understand and simulate the real world, they will bring new breakthroughs and changes to all aspects of human society.
04
Conclusion
There's a Chinese saying: "Sixes bring good fortune, but seven brings change." The number "seven" holds many mysteries, such as the seven-day cycle, the seven-day week, the seven-year itch, the seven-year hurdle for companies, and the ups and downs in an individual's career every seven years.
The 21st century has three sevens. One seven is already a headache, two sevens are even more complex, and three sevens represent "many," meaning many things are likely to change. The 21st century will be a very fast-paced period. Therefore, using 20th-century thinking may not be feasible. For example, in the digital age, we can leapfrog ahead and choose new paths, while simply copying Western ideas and management models may not work. Of course, this does not mean that our previous learning from the West was unnecessary; on the contrary, it was extremely necessary.
The emergence of Sora is a new milestone in human AI technology. We are currently lagging behind the United States in AI, but everything has its pros and cons, and the impact and disruptive consequences of AI on humanity have not yet been fully verified. One view holds that all creations and inventions are actually discoveries—that is, people "discovering" them at the right time. These patterns and scientific mysteries already existed and functioned before we discovered them.
The Tao Te Ching states, "The net of Heaven is vast; its meshes are wide, yet nothing escapes it." If we consider this "net" as an invisible and intangible high-dimensional "digital system," it connects and maps with the real world, recording and archiving everything in the real world. The advent of Sora allows for a seamless connection and integration between the real and digital worlds, and the interaction between the Net system, the real world, and the digital world will further bring about more changes to our future.
