Yang Jing sent from Aufei Temple
Qubit | Public Account QbitAI
Just now, the DAMO Academy’s top ten technological trends list was released.
This time includes basic technology fields such as AI, chips, and cloud computing. There are not only generative AI that has triggered a wave of global investment, but also integration of storage and computing on the eve of large-scale applications, and Chiplet design and packaging technology…
(Come and see first)
But different from the past, it can be seen only from the list that industry and integration have become the key words this time.
And in further conversations with the Bodhidharma Institute, there is a deeper underlying logic difference behind this.
How does DAMO see technological trends?
Continuing the previous practice, this trend list is divided into three dimensions: paradigm reset, industrial innovation, and scene change.
As mentioned above, in terms of technical dimensions, it includes basic technologies such as AI, chips, and cloud computing.
-
AI: multi-modal pre-training large model, generative AI;
-
Chip: Chiplet, integrated storage and computing;
-
Cloud Computing: cloud-native security, software-hardware fusion cloud computing architecture, and a predictable network with end-network integration.
And on this basis, the integration of various cross technologies has spawned new scenarios and new applications, such as dual-engine intelligent decision-making, computational optical imaging, and large-scale urban digital twins.
From the perspective of industry, it focuses on the field of ICT (Information and Communications Technology), which not only includes the two underlying technologies of computing (storage) and communication (network), but also horizontally includes the two major technologies of security (management) and application (industry). large application areas.
Compared with the past, this year’s technological trends given by Dharma Institute have a key word that cannot be ignored, that is integration. This is also the most talked about trend in conversations with Bodhidharma Academy.
There is the integration of computing and communication, hardware and software, digital technology and industry… the most obvious of which is the integration of AI technology and industries.
In this regard, Dharma Academy this year focused on the multi-modal pre-trained large model that is expected to reset the paradigm, and the generative AI that stimulates more scenarios.
Multi-modal pre-training large model
AI is developing from single-modal intelligence such as text, speech, and vision to general-purpose AI that integrates multiple modalities. This has been widely perceived in the past year.
Typical progress includes the image-text matching model CLIP proposed by OpenAI. The field of AI painting represented by the explosive Stable DIffusion basically revolves around CLIP.
There is also BEiT-3, Microsoft’s multimodal base model that “beats everyone at everything”. The model relies on only 1.9 billion parameters, only uses public data sets, and crazily brushes SOTA on 12 tasks, even including pure vision tasks.
Based on the vigorous development in the past year, it can be foreseen that, on the one hand, for individuals, it can stimulate creativity, especially for non-technical personnel, and no programming ability is required. On the other hand, the multimodal pre-training model will reshape the business model of artificial intelligence.
Technology companies with big data, computing resources, and model development capabilities will become model service providers, helping companies integrate basic model capabilities with production processes, thereby reducing costs and increasing efficiency.
Ali believes that establishing a unified, cross-scenario, multi-task multi-modal basic model will become one of the mainstream trends in AI development.
As an infrastructure in the future, large models will realize the unified knowledge representation of images, texts, and audio, and evolve towards cognitive intelligence capable of reasoning, answering questions, summarizing, and creating.
Generative AI
If the multi-modal pre-training large model is the future AI infrastructure, then generative AI is a scene innovation based on the technology base, covering graphic creation, code generation, games, advertising, artistic graphic design and other applications.
In the past year, the technical progress mainly includes three aspects.
-
Image generation: Diffusion models represented by DALL-E 2 and Stable Diffusion;
-
NLP (Natural Language Processing): ChatGPT based on GPT-3.5;
-
Code generation: Copilot based on CodeX.
With the explosive growth of content creation, the controllability of content quality and semantics is the current challenge. In terms of industrialization, there are still issues such as cost reduction, data security and controllability, and creative copyright that need to be resolved.
In the future, generative AI will become a popular basic technology, which will greatly promote the production and creation of digital content, and its application boundaries will also expand to more fields.
It can be seen that AI technology represented by multi-modal pre-training large models and generative AI has reached the eve of scale and the singularity before the outbreak.
On the one hand, human beings have begun to concretize the imagination of general artificial intelligence; on the other hand, this is also a major feature revealed by the top ten technological trends of DAMO Academy.
That is to pay attention to the industrial trends brought about by technological breakthroughs.
The same trend has also appeared in other technologies on the list. They have proven their industrial value and are waiting for market verification.
For example, the Chiplet chip packaging technology, which just established an alliance last year, will gradually unify its interconnection standards, which may restructure the chip R&D process; there is also a memory-computing integrated chip that breaks through Moore’s Law. Vertical segments ushered in large-scale commercial use.
However, it is not surprising that these trends exist in a world where cutting-edge technologies are advancing side by side and competing with each other. But after all, it is a prediction of the future. Why did Dharma Academy only choose these trends as the directions worthy of attention in the future?
It is necessary to mention the methodology of this time, which is also different from the lower level in the past.
Why do you see it that way?
Talking about the essence of methodology in this trend report and in the communication with Bodhidharma——
Back to Basic, starting from the basic field to gain insights into industrial changes.
As mentioned earlier, this time it is more focused than ever before—concentrating on the four major areas of ICT computing (storage), communication (network), security (management), and application (industry) for analysis.
Under the continuation of the previous analysis method of “quantitative divergence and qualitative convergence”, the factors considered comprehensively have also undergone slight changes.
This time, it is mainly based on factors such as theoretical attention, technical feasibility, degree of industrialization, and social value.
From theory to technology, to industrialization and market verification of social value, this is precisely the basic path for cutting-edge technology to move from scientific theory to the market and bring inclusive value.
In the past, taking last year’s trend list as an example, their selection dimensions paid more attention to factors such as technical feasibility, social value, and theoretical attention.
Underneath the superficial phenomena, Qin Xi, a member of the Organizing Committee of the Top Ten Technological Trends of DAMO Academy, mentioned a key issue:
Whether some very good scientific theories can form technology needs to pass three major tests.
The first test is the technology itself, whether there is an increase in magnitude, and what kind of effect it will bring; the second hurdle is the industry, whether it can form a complete upstream and downstream industrial ecology, and whether it can be engineered and productized; the third is It is a market test, can it be accepted by consumers? Does anyone pay for the technology.
In the 1990s, in “Basic Science and Technological Innovation: Pasteur’s Quadrant”, Donald Stokes divided technology from theoretical research to application into four quadrants, Bohr’s Quadrant, Pasteur’s Quadrant, Edison’s Quadrant, and Peterson’s Quadrant.
Among them, the research idea centered on Pasteur’s quadrant is the path from theory to technology use.
This is also the entry point of this research on the top ten technological trends of Dharma Institute. More specifically, it is about to be verified by the market and is expected to form the mainstream trend of the industrial ecology.
Based on this idea, they also revealed in this study that they had to make choices about many technological trends. For example, in terms of security, they did not choose quantum encryption and trusted privacy computing, which had also caused heated discussions in the past, and finally chose cloud-native security. .
Many experts reflected in the survey that although network security is a follow-up technology, many technologies come first.
For example, if there is no cryptographic technology to deal with the “post-quantum era” in advance and respond to quantum computing, then many problems will erupt at the same time in the future.
However, in the past year, whether the underlying technology has made significant progress in the comprehensive field, and whether the entire industry is niche or not, the organizing committee finally made a strict screening.
Why is it worth paying attention to?
Finally, back to the technology trend itself, there are too many organizations that provide forecasts every year. Why is the list given by Dharma Institute worthy of attention?
First of all, it is forward-looking and is the vane of technological development every year. This year is the fifth time that Bodhidharma has released the trend list. Before that, it has successfully predicted many fields such as AI for Science, large-scale models, and AI pharmaceuticals.
Especially the AI for Science predicted last year, it can be seen that in the past period of time, AI for Science technological achievements have exploded intensively, playing an increasingly important role in biomedicine, materials, physics, chemistry, and even mathematics.
Secondly, it is also professional. From the annual selection process, we can see that in the process of the birth of the top ten trends of Dharma Academy, the big data of papers and patents will be “quantitatively diverged” first, and then the fields of production, learning, research, and application will be “quantitatively diverged”. In-depth interviews with nearly a hundred experts “qualitative convergence”, cross-validation from interviews with scientists, entrepreneurs and engineers.
Wu Hequan, an academician of the Chinese Academy of Engineering who participated in the survey and interview, said that the DAMO Academy’s Top Ten Technology Trend Report will serve as a good reference for the entire ICT industry, including technological innovation and ICT application research in vertical industries.
Finally, from the perspective of returning to the essence of industry this time, the integration of cutting-edge science and industrial technology is far closer than ever. This happens to be the externalization of the current machine revolution.
If the time line is stretched, in fact every industrial revolution is related to the fusion relationship between the two.
Qin Xi, a member of the Top 10 Technology Trends project team of DAMO Institute, gave an example: From the perspective of industry, the first industrial revolution, science and technology are still relatively far away; the second industrial revolution, technology and science began to be closely related. , science began to guide technology slowly; it became more obvious in the third information revolution.
If the paradigm standard of the second industrial revolution was scale and standardization, then the third industrial revolution is the unification of individualization and scale.
There is no doubt that we are in a whole new technology cycle. Comprehensive and in-depth integration with the industry has become the meaning of the title.
What kind of technological trends are worth paying attention to in the new year? Dharma Institute gave an answer.
(Disclaimer: This article only represents the author’s point of view, not the position of Sina.com.)
This article is reproduced from: https://finance.sina.com.cn/tech/csj/2023-01-11/doc-imxzuxpn2177459.shtml
This site is only for collection, and the copyright belongs to the original author.