To prevent getting lost, the elevator goes directly to the safety islandNewspaper man Liu YaEast A
Youbin
Recently, OpenAI announced the artificial intelligence ** generation model SORA, which caused a sensation all over the world, SORA can directly generate realistic ** according to the text instructions entered by the user, and the fluency and logic can reach the level of fake.
The advent of SORA shows that AI technology has advanced to a new stage, not only in the production of **, such as film, advertising industry direct application, but more importantly, once AI can generate smooth and logical **, it can be more accurate identification. It is conceivable that self-driving cars can directly judge the situation on the road with the help of AI in high-definition ** that can be captured by cameras, just like human driving, which is completely different from the existing technical path; The surgical robot can directly judge the location of the lesion and the scope of the operation with the help of the AI system, without relying on the doctor's experience and technology. These are very difficult to implement under the existing technical framework, requiring a large amount of data labeling, complex algorithms and training, and the results are not necessarily very good.
Nearly at the same time as OpenAI released Sora, Google and Facebook's parent company, Meta, also released similar products. The competition of Internet giants in this field has reached a white heat, mainly to see the prospect of related technologies, which is not only the progress in the field of production, but also the direction of general large models. Perhaps, in the near future, all electronic devices may be installed with general large models, mobile phones, cars, medical devices, etc., AI large models are more like operating systems Windows, Linux, Android, become the necessary underlying system of the device. This cake is so big that the industry can't imagine it at present, so it is understandable that the major giants are fiercely competing.
Many people lament that the gap between AI technology in China and foreign countries may gradually widen, and the next product similar to the operating system level may still be released by companies such as OpenAI, Google, and Meta. The reasons for this are very complicated, and there have been a lot of such voices since ** recentlyThey believe that the gap in this field in China has widened because the AI technology in the United States is no longer open source, so domestic companies can no longer plagiarize.
This view is patently wrong and a misconception of open source software technology.
First of all, the open source of the AI field has not decreased, on the contrary, there are more AI-related open source models. Even for large models, there are open sources from Google and Meta, and the basic Transformer architecture is even more open source, and anyone can develop on top of it. But open source doesn't mean that all technical secrets will be made public, and each company will have its own way of processing data, model training methods, and fine-tuning algorithms. There has been no significant change in foreign policies on open source technology, and it is unreasonable to say that the gap in the AI field has widened because the United States is no longer open source.
Second, open source software is not a charity, but more of a business strategy. Open source software is not exactly the same as free software, and many enterprises do not simply contribute to the industry, but a more clever business strategy. For example, the well-known development language J**A and the database system MySQL are both open source, but they are worth a lot of money, and the purchase price is billions of dollars. Enterprises with open source software can quickly attract traffic, save development costs, and marginalize competitors through open source projects, while launching commercial versions and value-added services on the basis of open source projects. Android is Google's open-source operating system, but Google obtains a lot of commercial benefits every year through Android, and Android is the main entrance to Google's applications.
On the one hand, it is to contribute traffic to open source software, purchase value-added services, and at the same time, it also needs to contribute to the development of open source software. To some extent, using open source software is equivalent to thinking of yourself as a potential customer of the owner of open source software, and you are a user who will contribute technology, so the use of open source software cannot be plagiarized.
Third, the gap between Chinese and foreign AI tends to widen, due to many reasons, but it has nothing to do with whether it is open source or not. The main reasons that can be thought of are talent reserves, data resources, hardware resources, industry ecology, resource investment, and policy environment. Although China also has a lot of talents in the field of AI, in comparison, in terms of generative AI, American companies and universities have been involved in this field for a long time, and the author produced an analysis report in 2019 showing that in fact, before 2018, Google, Microsoft, and Apple have applied for many patents for generative AI, and in comparison, China's generative AI patent applications are mainly after 2023. In the field of generative AI, computing power is particularly important, and the limitations of GPUs also have a certain impact, many model training requires more than 4090 resources, especially NVIDIA's chips, which already have a good ecosystem, and even if the hardware can be replaced in a short time, it is difficult to establish a similar software ecology. At the same time, the data of model training is also particularly important, and the quantity and quality of Chinese data are far from English, which itself has a great impact. In addition, the regulations of the policy environment also have an impact, and large language models need to be tolerant of the accuracy of generative content. These are all questions worth pondering.
In short, there are many reasons why the gap in the AI field tends to widen, but it has nothing to do with whether it is open source or not.