laitimes

Chief Outlook|Wu Chao of China Securities Construction Investment: AI technology is expected to usher in leapfrog development in the Year of the Dragon

Chief Outlook|Wu Chao of China Securities Construction Investment: AI technology is expected to usher in leapfrog development in the Year of the Dragon

What are the trends of artificial intelligence (AI) in 2024, how will the industry evolve, and which segments will benefit the most?

Recently, The Paper launched the "Spring Breeze Blowing and Born" - "Chief Wired" 2024 New Year's Eve Market Outlook, interviewing a number of chief economists, chief analysts, and star fund managers, to take the pulse of the new main line of investment in the Year of the Dragon, explore new market opportunities, and look forward to new market trends.

This issue is an exclusive interview with Wu Chaoze, director of the China Securities Research Institute and administrative head of the international business department.

From 2013 to 2020, he ranked first in the communications industry for eight consecutive New Fortune Best Analysts, focusing on research in the fields of 5G, cloud computing, and the Internet of Things. He is currently a member of the Securities Analyst, Investment Advisor and Chief Economist Committee of the Securities Association of China.

Chief Outlook|Wu Chao of China Securities Construction Investment: AI technology is expected to usher in leapfrog development in the Year of the Dragon

AI technology is expected to develop by leaps and bounds

"In 2024, AI technology is expected to usher in leapfrog development. Wu Chao said that with the breakthrough of the underlying algorithm, the ability of large models will continue to improve.

Wu Chao said that in general, the large model has four stages of development: the first stage is "knowing language", that is, letting the large model learn language, reaching the level of human language understanding, generation, and logical reasoning, and the second stage is "knowing tools", that is, letting the large model learn to use tools, such as realizing takeaway reservations through artificial intelligence.

"The third stage is 'planning and decision-making', that is, allowing the large model to have stronger logical reasoning ability and realize the implementation of planning and decision-making, and the fourth stage is 'emotional and conscious', that is, to let the large model have emotion and awareness. Wu Chao pointed out.

Wu Chao further pointed out that on the whole, the large model has now begun the second phase. OpenAI allows ChatGPT to access the internet and use hundreds of third-party plug-ins, which cover a comprehensive range of scenarios such as tools, travel, life, games, and learning.

"In layman's terms, GPT-4 has gradually become an 'expert' from an 'encyclopedia'. The GPT-3.5 model has shown good performance in some relatively simple human professional tests, but when the complexity of the task increases, it becomes difficult to handle, and the GPT-4 model has achieved quite good performance in some relatively difficult tasks. Wu Chao said.

Wu Chao said that due to the breakthrough of the underlying algorithm technology, as well as the good support of computing power and data, the singularity of the large model has arrived.

It is expected that the domestic AI explosion in the second and third quarters is expected to appear intensively

In terms of application, Wu Chao said that from text interaction to pictures, audio, video, 3D and even feedback in the physical world, large models at home and abroad continue to broaden the types of modalities that can be identified or generated, enrich the way users can interact with AI, and give birth to popular applications such as photo application Miao Ya Camera, AI video application Runway and Pika.

"In the future, it is expected that the transition between modalities will be smoother, which will unlock more application scenarios and product forms. Wu Chao said.

In terms of domestic, Wu Chao said that the leading large models in China have basically achieved a breakthrough in the capability boundary from October to November 2023, and achieved the same or even some capabilities to surpass ChatGPT.

"Therefore, by analogy with overseas, with the iteration of domestic large-scale model capabilities, in the second and third quarters of 2024, it is expected that AI explosive applications will appear intensively. Wu Chao further pointed out, "GPT-4 will be released in March 2023 and will lead the iteration of model technology in the follow-up, and the landing of overseas AI explosions will also be roughly 2 to 3 quarters after the release of GPT-4, and the application field has achieved a hundred flowers." ”

The computing power on the device side is worth looking forward to

In terms of computing power, Wu Chao said that the rapid iteration of chips and the strong demand for AI computing power make it worth looking forward to on the end of the device. As an important supplement to cloud computing power, device-side AI computing power has multiple advantages under the current trend of cloud-side AI transformation to device-side AI.

First, the independent operation of terminals and the collaborative workload of cloud and terminal can greatly reduce the computing power demand and energy consumption cost of the cloud.

The second is to solve the problems of use in extreme scenarios such as "weak network" and "overflowing clouds" to ensure low latency and high reliability.

The third is strong privacy protection and strong customization, which can generate proprietary user portraits.

Fourth, it has the natural advantage of natural language voice interaction, which can easily obtain pictures, photos, videos, locations and other information to optimize the user experience.

"Starting in 2022, optical modules in the traditional cloud computing market in North America have begun to upgrade to 800G speeds. In 2023, as the demand for GPU computing power for high-quality networks for AI large model training increases, 800G optical modules will begin to be shipped in batches, and shipments are expected to increase significantly in 2024. Wu Chao said.

Wu Chao pointed out that increasing network bandwidth can significantly improve computing power and is very cost-effective, and GH200 is an important example. Among them, technologies such as silicon photonics and thin-film lithium niobate have the advantages of low power consumption, low cost and high integration, and the chances of achieving breakthroughs have been greatly improved.

"Domestic optical module companies are in the world's first-class position, the layout of optical devices maintains a leading position in the world, and EML, silicon photonics and thin-film lithium niobate chips also have the possibility of continuous breakthroughs. Wu Chao further pointed out.

Abundant application scenarios

At the industrial level, Wu Chao believes that as the whole industry changes from informatization and networking to digitalization and intelligence, the digital economy will become the main economic form after the agricultural economy and the industrial economy.

"On the whole, the digital economy is mainly divided into three stages: one is the PC stage, the second is the mobile Internet stage, and the third is the intelligent era. Wu Chao further pointed out that the core characteristics of the digital economy are "data" and "algorithms", and at the same time, digitalization is two-way, and the vast majority of enterprises in the future are digital enterprises, and intelligence is the ultimate goal.

Wu Chao said that AI, as a phenomenal tool to accelerate the implementation of digital applications, is the "operating system" of the digital era. ”

In terms of scenarios, Wu Chao pointed out that the Al model will bring revolutionary changes to the financial industry, and the digital demand of the financial industry is rigid and the investment is huge, which is a big stage for the application of the large model.

At the same time, the medical field is also one of the important scenarios for the comprehensive empowerment of AI large models. On the one hand, common scenarios such as consultation, medical record generation, and patient history analysis require doctors to integrate, summarize, analyze and judge based on historical information, which is what large models are good at, and on the other hand, AI large models also have broad application fields in the fields of disease prediction, drug research and development, clinical trial optimization, and precision medicine. Through AI-enabled medical care, some grassroots hospitals are expected to achieve an all-round improvement in medical standards. Wu Chao said.

Wu Chao further pointed out that the combination of industry and AI large models will promote the intelligent upgrading of the industry, improve production efficiency, reduce costs, and achieve Industry 4.0. Through deep learning, AI large models optimize production processes, realize intelligent manufacturing, and help industrial development. This convergence will lead to innovations and breakthroughs in the industrial sector and promote sustained economic growth.

Pan-AI thematic investment is expected to continue to evolve

On the investment side, Wu Chao said that in 2024, pan-AI theme investment is expected to continue to be interpreted and enter the stage of removing the fake and retaining the true.

"In 2023, the structural situation of the A-share market is more obvious, especially the TMT direction of the sector has shown a strong momentum, looking forward to 2024, around AI artificial intelligence, there are three clues worthy of special attention by investors. Wu Chao pointed out.

One is the large model. Wu Chao believes that large models occupy a core position in the AI industry chain and will be the most valuable part of the industry chain in the future. Especially after OpenAI launched the plug-in function, the positioning of the large model in the entire ecosystem is similar to the operating system + app store in the smartphone era, and the leading company in this link is expected to have a trillion-level market value. Therefore, the focus is on the localization of the large model technology base, such as EDA and other subdivisions.

The second is computing infrastructure. Wu Chao said that as artificial intelligence enters the era of large models with huge parameters, the growing demand for computing power has ushered in huge opportunities for the AI chip and server market, and it is optimistic about the domestic computing power chip and server-related industrial chain in the field of computing power. Represented by AI chips, they also include ICT equipment, optical modules, servers, etc., which are the "water sellers" in the AI ecosystem, or similar to the "lithium mines" in the new energy vehicle industry chain, with traceable performance and the most deterministic link in computing power.

The third is application. Wu Chao expects that the landing and application of AI in vertical fields will be the main line in 2024, and is optimistic about the landing of AI in education, autonomous driving, end-side equipment, and industrial scenarios. From to C to B, the current plate effect is graphic, office, games, e-commerce, marketing, video, etc. In the short term, overseas applications that can access ChatGPT are the most direct benefits, and some companies have begun to reflect the value of AI to reduce costs and increase efficiency.