Recently, the "AI 'Xing' Vision Salon" hosted by ZTE was held in Beijing, where Cui Li, Chief Development Officer of ZTE, and a number of industry leaders focused on the past and present of artificial intelligence technology development, the latest progress, application trends, new paths of industrial integration and other hot topics for in-depth exchanges.
1. The general trend of digital and intelligent economy is the same, and opportunities and challenges coexist
Cui Li said that at present, we are in a new era of digital intelligence, and the deterministic trend of digitalization, networking, intelligence and low-carbon is irreversible, which not only brings us a huge market space, but also makes a positive contribution to the sustainable development of the global economy. However, any opportunity comes with challenges, and just like two sides of the same coin, we must think carefully about how to deal with the relevant challenges and avoid potential risks while seizing the opportunities.
In recent years, the global economy has been affected by the epidemic and other factors, the pressure on recovery has increased, and countries have also produced a series of chain reactions in the process of solving their own problems.
With the acceleration of digitalization and intelligence, information explosion and information scarcity coexist, information authenticity has become a major challenge, and digital ethics issues are becoming increasingly prominent, such as AI face swapping technology being used to defraud and mislead the public.
With the deepening of digitalization and networking, the amount of data has surged, and the demand for computing resources has also increased, especially after the introduction of AI technology, the demand for computing power and energy has reached an unprecedented height. Therefore, how to maximize output, improve resource utilization efficiency, and accelerate monetization under the constraints of existing resources has become a key issue that we need to consider.
2. Be realistic and pragmatic, be inclusive, and promote the healthy and benevolent development of AI
Cui Li reviewed the past and present life of AI development, she said, looking back at the development of artificial intelligence technology for about 70 years, there have been two springs and two cold winters, and at present, we are in the third spring, behind every key breakthrough in AI, scientists and industry leaders have been working hard for several years or even decades, "all the things that have come out of nowhere are actually planned for a long time." Even ChatGPT is not accidental, it is also based on previous generations of GPT, and it is also based on the joint efforts of data, algorithm optimization, and computing power enhancement. Looking back at the past two AI winters, the most important reason is excessive disappointment caused by high expectations, this time, as spring comes, we need to look at AI more pragmatically and calmly to help it develop healthily.
She said that at present, although people rely on Transformer to build a new AI paradigm, it still does not escape the limitations of mathematics, people use Transformer as the main means of neural networks to capture the hidden knowledge paradigm in massive human data, when the model encounters new problems, the use of the knowledge paradigm mastered in the early stage to make reasonable inferences about the new situation through extrapolation or interpolation. In Cui Li's view, the biggest contribution of Transformer in this wave lies in two points: first, generalization, Transformer can not only have strong capabilities in a single field by summarizing massive knowledge, but also widely empower all walks of life as a general basic technology; The second is the emergence, quantitative change to qualitative change, the emergence of capabilities enables AI to deal with problems that have not been encountered in the training stage, and through long thinking chains and other means, multiple queries of internal knowledge can enable it to solve more complex problems.
However, she also mentioned that in this process, there are several aspects that need to be paid attention to: first, on the issue of scale, although "vigorously produce miracles", for subsequent development, relying solely on the way of scale stacking will bring a waste of resources; The second is the expansion of capabilities, from unimodal to multimodal, and then from simple to complex problem processing, the knowledge density of the model will continue to increase, and the knowledge density is the source of value. The ability to achieve higher efficiency and higher value is likely to become the core direction pursued by the industry.
To this end, Cui Li put forward her thoughts on how to help the healthy, benign and benign development of AI in the future:
She emphasized that GenAI is still in the early stage of development, with rapid iteration of technology, confusing market, and also generating issues such as illusions, privacy, security, and ethics. At present, industry self-discipline is particularly important, leading enterprises should first self-restraint, such as setting up a responsibility framework and self-discipline mechanism, and constantly implement these principles in the process of AI model training and product development and application, and at the same time work together to continue to innovate and improve, solve problems, and make AI more healthy.
AI is a highly interdisciplinary field that requires deep integration and efficient collaboration with various technologies such as big data, computing power and networks, material technology, embodied intelligence, and energy transition. In addition, from the perspective of ethics and security, it is necessary to ensure that the design and application of AI conform to universal human moral standards and values, such as equality, inclusiveness, and goodness, minimize bias and discrimination, and continuously optimize the diversity of data and the fairness of algorithms, etc., which require the combination of humanities and sciences, including ethics, law, sociology, etc.
Based on this, Cui Li put forward some judgments: first, open source and closed source have their own advantages and disadvantages, and it has become an inevitable trend for the two to coexist and promote each other; Secondly, in the pursuit of scale, the size needs to be determined according to the specific application scenarios, problems, goals and resource constraints to achieve the best input-output ratio. Thirdly, it is difficult for us to take into account both breadth and depth at the same time, and we should analyze the specific problems of the scene in combination, sometimes a small system with multi-module collaboration may have better performance than a single large system, in fact, practice has also proved that the MOE of model training, the agents of model application, and the combination of large and small models have better cost performance; She also said that AI is silicon-based, inorganic, and mathematical, while human beings are carbon-based, organic organisms, in addition to natural sciences and social sciences, etc., its cognition, decision-making and behavior have a profound cultural, historical, values and other backgrounds, at the same time, at least for now, AI is not conscious, we should not over-anthropomorphize it, and do not over-stress the new problems brought by AI, just deal with them objectively and rationally, and the process of human science, science and technology and social progress is itself a process of discovering and solving problems; Finally, the integration of the digital world and the physical world is deepening, but the complex factors in human society, such as economy and politics, cannot be programmed purely by algorithms, and more humanistic spirit must be introduced. In addition, high-quality data will be an important cornerstone to drive the development of AI in the future.
3. Empower AI advancement in response to the situation by focusing on network computing, open decoupling, and training and promotion
Based on the above thinking, Cui Li introduced ZTE's three core propositions focusing on key challenges: strong computing with the network, openness and decoupling, and simultaneous training and promotion. In terms of network power computing, whether it is the bare die interconnection within the chip, between chips, between servers, or between DCs, the continuous innovation and breakthrough of network connection technology will greatly improve the performance and efficiency of intelligent computing. Therefore, "computing with network strength" has become ZTE's core strategy; To this end, ZTE continues to promote the decoupling of software and hardware, training and deduction, and model, promote the componentization and sharing of various capabilities, break down technical barriers, accelerate the innovation, R&D, application and commercialization of AI technology, and promote an open technology ecosystem. She firmly believes that the value of AI is not limited to general artificial intelligence, and at the same time, for social development, improving production efficiency is much higher than giving emotional value, so the simultaneous development of training and promotion is the key to the take-off of the real economy driven by AI. To this end, ZTE not only applies the R&D model and telecom model in its own field and builds a data flywheel, but also carries out practice in many fields such as water conservancy, urban lifeline, industry, and park security.
To this end, ZTE will position itself as an end-to-end intelligent computing solution provider, and internally determine intelligent computing as the company's long-term strategic main channel, and comprehensively start the expansion of intelligent computing through the way of "chip + complete machine + assembled R&D + large model", and will rely on the company's long-term deep cultivation in the field of communications, assembly R&D, software, hardware and system engineering practice. We will work with industry partners to promote innovation and breakthroughs in the industry.
In addition, the salon also set up an interactive Q&A session, in which guests from all parties discussed in depth the feasibility of heterogeneous computing power, the commercialization path of AI, the application and development of AI in the industrial field, the construction trend of domestic intelligent computing centers, the development prospects of small models, and the integration of computing power.
As one of the industrial ecological activities specially organized by ZTE, this salon is a useful discussion on the future development of artificial intelligence, facing the future, ZTE will build an information communication bridge with customers, partners, industry experts, opinion leaders, and media through a series of ecological activities such as the salon, and continue to build and expand the ICT industry innovation ecosystem with an open, integrated, and win-win mentality, and work together to win a new era of digital intelligence.
Note: This article does not constitute any investment advice. The stock market is risky, and you need to be cautious when entering the market. There is no harm in buying and selling.
![Let the local people drink "safe water" - the 31st trip of Chinese scientists to Sri Lanka[fig]](data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=)