Technology cloud report original.
ChatGPT has been soaring all the way. Behind the IQ of SAT710 is not only the "brain" of coeloid animals, but also the "central nervous system" of cloud computing, which has continuously developed into the "digital native intelligence" of vertebrates.
In the more than ten years since cloud computing entered China, its essence has undergone fundamental changes. It has realized the central nervous system evolution from the initial virtualization technology to the digital world, and completed the sublimation from service-oriented resource delivery to cloud-native value empowerment. .
However, most enterprises' understanding of cloud computing has not kept up with the pace of the times, and they have not given cloud computing a "digital social status" that rivals its central nervous system value in digital transformation.
The explosion of ChatGPT objectively demonstrates the central nervous function played by cloud computing in the leapfrog evolution of artificial intelligence, and officially opened the prelude to re-understanding cloud computing.
The future has come, and in the digital universe, what is the role of cloud computing? How much is it worth? How should enterprises internalize the value of cloud computing and complete the evolution to a new digital native entity?
Recently, Li Wei, deputy director of the Cloud Computing and Big Data Research Institute of China Academy of Information and Communications Technology, accepted an exclusive interview with Science and Technology Cloud Report and shared her unique insights on cloud computing.
Li Wei, deputy director of the Institute of Cloud Computing and Big Data, China Academy of Information and Communications Technology
The success of ChatGPT is also the success of cloud computing
In the era of digital natives, ChatGPT is a product with quite the characteristics of the times.
As a phenomenal AI tool, almost everyone has turned their attention to the algorithm level of ChatGPT.
But in Li Wei's view, cloud computing plays a very critical role in it, which is as important as algorithm optimization.
"The marriage of OpenAI and cloud computing is like the brain finding the central nervous system. It not only spawns explosive products like ChatGPT, but also fully releases the competitiveness of digital life forms built with cloud computing as the central nervous system." Li Wei explain.
Throughout the development history of application models in the past ten years, artificial intelligence models continue to emerge, but there has never been an intelligent application that has changed human social life as profoundly and extensively as ChatGPT. Too many companies focus on the optimization and improvement of model algorithms and ignore The central nervous system of cloud computing may be one of the reasons.
According to Li Wei's analysis, ChatGPT is a "digital-native" artificial intelligence, and cloud-native concepts and technologies are important factors for its success.
First, cloud-native technology provides ChatGPT with a unified scheduling and distribution service capability for ultra-large-scale parallel computing, which significantly reduces ChatGPT training costs.
In the training of ChatGPT, as many as 175 billion parameters are calculated at the bottom layer. This kind of ultra-large-scale model calculation requires the unified arrangement and scheduling of the large-scale computing power at the bottom layer. However, it is difficult to achieve this capability simply by relying on GPU to provide computing power. , requires cloud computing technology to realize distributed computing.
This point of view has also been confirmed in the reply of ChatGPT.
ChatGPT said that it uses distributed computing in model training. The underlying cloud-native K8S service is distributed into super-high-level node scheduling according to different dimensions for distributed node computing.
When asked how many distributed computing nodes there are, ChatGPT said that the scale is too large to disclose.
This also reflects to a certain extent that the training of large models consumes a lot of resources, and the cost of training a product like ChatGPT is very high. Li Wei said that without the support of cloud computing, time costs and economic costs will rise in both directions.
According to public information, in the past two and a half years, the training cost of large models such as GPT-3 has dropped by more than 80%, from the first training cost of 12 million US dollars to 1.4 million US dollars. A very important reason is the adoption of cloud computing technology ——Multi-dimensional information deployment, distributed scheduling calculation, and data separation and mixing technology have greatly saved the training cost of AI models.
The second is that cloud native technology can realize multi-dimensional application deployment, realize ChatGPT's situational learning mode and multi-modal (text, code, picture and other multi-dimensional information) input.
ChatGPT was able to break through the previous artificial intelligence that could only focus on one-dimensional training such as text, pictures, and voice, and turned to multi-dimensional training, bursting out amazing natural language processing capabilities. The important reason is cloud native technology.
Among them, the cloud-native container, as the neuron of the digital life body, carries the model parameter type and the container node of the scheduling connection, has the ability to standardize the deployment of multi-dimensional applications, and standardizes the access and deployment of heterogeneous information such as text, voice, and images. Eyes, ears, nose, tongue, body, and mind" multi-dimensional information deployment and processing accelerates the evolution of the digital brain, that is, the artificial intelligence algorithm.
The third is to realize discrete training and online fine-tuning through cloud-native off-hybrid technology, and realize the fine-tuning training mechanism and prompt-guided mode of ChatGPT's Human Feedback Reinforcement Learning (RLHF).
Based on a large amount of high-quality data and corpus training, ChatGPT has achieved breakthroughs in dialogue intent recognition and content generation capabilities, mainly because ChatGPT has a powerful foundation for intelligent algorithm learning and memory recall.
Through the cloud-native offline hybrid and extremely flexible call mechanism, offline training of hundreds of billions of ultra-large-scale parameters has formed the storage and memory resource pool of ChatGPT, and completed the fine-tuning training mechanism and prompt guidance of human feedback reinforcement learning (RLHF) through online supplementation mode, simulating the feedback of each brain functional area, and realizing self-supervised learning under small-scale feedback of the relationship between multi-task words, sentences, and pictures.
Fourth, cloud-native + AI-native linkage opens up a new business model and forms a virtuous ecological cycle.
On the development side, ChatGPT relies on cloud computing services. Over the years, OpenAI has received tens of billions of investments, which help OpenAI run and train its models on the platform; on the product side, OpenAI develops applications based on Cloud Native. The convenient high-performance computing computing model and polishing algorithm provided by computing, and external sales of products and APIs; and investors based on AI Native to improve search, painting and other products, will be used in Office, Teams and even operating systems in the future to provide OpenAI Ability.
Cloud computing: the central nervous system of digital life
"Traditional artificial intelligence focuses on the innovation of model algorithms, but ignores that intelligence is a complex systematic project. Just like human intelligence, overall intelligence should include parts of the brain area and the central nervous system. Intelligence with only the brain area and no central nervous system is static. Artificial intelligence should be based on a perfect central nervous system."
This metaphor also allowed Li Wei to find the unique positioning of cloud computing - cloud computing is the central nervous system of digital life.
First of all, the central nervous system of cloud computing can realize the scheduling system of linking computing power and brain regions (applications) in the digital world.
Distributed cloud computing can schedule ultra-high-volume heterogeneous high-performance computing GPU and general-purpose computing CPU resources. For heterogeneous computing resources, cloud computing shields the complex deployment framework and provides a standardized computing environment for large-scale parameters, which can be used efficiently Heterogeneous computing power reduces a lot of development overhead and operating costs, and improves efficiency.
Secondly, the central nervous system of cloud computing can realize the access, deployment and connection of multi-dimensional brain regions (multi-dimensional applications) in the digital world.
Most traditional artificial intelligence is like early coelomates with only one brain area, and can only learn and train single-dimensional parameters, while ChatGPT can learn multi-dimensional (image, voice, text, etc.) information around a certain topic at the same time, because of cloud computing It can realize the standardized deployment of multi-dimensional model parameters, so that cloud computing can standardize the deployment of different application parameters on different nodes through technologies such as cloud-native containers, and connect them to each other for processing, and realize the unified processing of multi-dimensional data through cloud-native databases.
Just like the central nervous system of the human brain, it connects brain areas that process different levels and dimensions.
Thirdly, the central nervous system of cloud computing can realize the storage and memory of the digital world (offline training) and various brain regions (online prompt guidance).
ChatGPT offline training is supported through cloud-native offline hybrid technology, and the online prompt-guided algorithm is realized, just like the human central nervous system will continuously transmit experience and real information to the brain for brain information correction. From the user's point of view, ChatGPT can adjust the answer according to the communication, which is very intelligent.
"For so many years, everyone thinks that the cloud is very important, but it is not clear what cloud computing is for and what role the cloud plays in the digital world. It is very similar to the central nervous system, and it is not clear where it is. But it is very important. In fact, cloud computing is as ubiquitous as the central nervous system," said Li Wei.
Digital native enterprises give priority to building a "central nervous system"
"The significance of cloud computing as the central nervous system of enterprises should be more widely recognized."
In Li Wei's view, although the domestic cloud computing market has developed for many years, most companies still do not realize the importance of cloud computing.
Many enterprises are still stuck in the early cloud service era, using cloud computing as virtualized resources, and only see the value of cloud computing in reducing costs and efficient development.
But in fact, cloud computing is the central nervous system of an enterprise, which can realize business software integration, data integration processing, and process IT reengineering among various departments of an enterprise.
In fact, just because most enterprises do not realize the deep value of cloud computing, the domestic cloud access rate has been stagnant.
According to statistics from the my country Academy of Information and Communications Technology, in 2021, only 30% of Chinese companies will go to the cloud, and traditional industries such as industry, transportation, and energy will have an even lower rate of going to the cloud, only about 20%. This figure has not changed significantly in the past two years.
As the global digitalization process deepens, digital natives will become the core competitiveness of enterprises. Li Wei said that due to the development of the central nervous system, digital native enterprises are naturally adapted to the rules of the digital world and have strong competitiveness in the digital world, leaving traditional enterprises far behind.
"Just as cloud computing realizes ChatGPT's comprehensive multi-dimensional information processing and integrated scheduling of heterogeneous computing power resources, cloud computing, as the central nervous system of an enterprise, can open up the entire process business chain from upstream procurement to downstream customers, break the chimney process, and realize Purchasing, production, sales, channels, user consumption, personnel, property, and other full-business process reengineering and innovation, as well as unified aggregation and processing of data in all links.” Li Wei said that cloud computing combined with artificial intelligence, big data and other technologies can help enterprises become complete intelligent bodies with "central nervous system" and "brain", that is, transform and upgrade into "digital native new entities"
In Li Wei's view, digital native enterprises have obvious characteristics: digital thinking such as cloud native and big data is carved into the DNA of the enterprise, and they have the four-in-one intelligence ability of central nervous system, brain, open learning ability and intelligent innovation. The evolution process of these enterprises' digital natives is developed based on cloud native, the most important neuron capability support technology.
All enterprises hope to achieve success in digital transformation, but if enterprises want to win in the digital native era, the significance of cloud computing as the central nervous system of enterprises should be widely recognized. Not only that, Li Wei believes that more assistance needs to be given to enterprises.
One is to create higher-quality, cloud-native products to meet the needs of enterprises in all aspects and at multiple levels.
Li Wei suggested that cloud service providers provide subdivided products and service solutions for enterprises of different industries, sizes, and types. For most small and medium-sized enterprises, provide the central nervous system with "plug-and-play" capabilities; for large enterprises, establish a "central nervous system"; for traditional enterprises and other enterprises that have just migrated to the cloud, plan for the central nervous system from the very beginning .
The second is to improve the standards and evaluation system, and guide enterprises to establish a "central nervous system" and transform into digital natives.
At present, China Academy of Information and Communications Technology has released the "Cloud Native Technology Architecture Maturity" standard to help enterprises build a mature cloud native technology architecture; "Enterprise Digital Maturity Model IOMM" formulates an overall perspective for enterprises in different industries and sizes And the maturity model from the perspective of the enterprise IT department, through the "six major capabilities + six major values + five maturity levels" to evaluate the level of enterprise digital transformation capabilities, to help various enterprises to better build their own "central nervous system".
Although the digitalization level of most domestic enterprises is still far from the "central nervous system", Li Wei is not discouraged.
"In the digital age, cloud computing is becoming the central nervous system of digital life forms and the birthplace of digital transformation for enterprises. This trend has arrived. What we need to do is to explore the development path of new digital-native entities, promote relevant experience, and let more Traditional enterprises understand the concepts of 'digital native' and 'central nervous system', and continue to enhance their innovation vitality and competitiveness."
In Li Wei's view, cloud computing technology will continue to evolve in the future: first, redefine software and hardware to achieve unified scheduling and scheduling of heterogeneous computing power; second, create an operating system that integrates computing, network and cloud; Interface, through the standardized API interface, create neurons all over the digital world.
By then, the value of cloud computing as the "central nerve" will empower thousands of industries, and more enterprises will become new digital-native entities growing in the digital world.
[About Science and Technology Cloud Report]
Focus on original enterprise-level content experts - technology cloud reports. Founded in 2015, it is the top 10 media in the cutting-edge enterprise IT field. Recognized by the Ministry of Industry and Information Technology, Trusted Cloud, one of the official media designated by the Global Cloud Computing Conference. In-depth original reports on cloud computing, big data, artificial intelligence, blockchain and other fields.