Since the beginning of this year, ChatGPT has become more popular than expected, triggering a global discussion about the scientific and technological revolution, and at the same time, the topic of artificial intelligence has once again aroused people's heated discussions. On May 25, during the 2023 Xiaomanwaist Technology Conference, Dr. Zhang Jun, chairman of CEIBS Capital and former vice president of Huawei, said in an interview with Aoyi News that the advent of ChatGPT has made a milestone leap in artificial intelligence technology and pushed the concept of metaverse a step forward.
Zhang Jun believes that the gap between domestic AI models and ChatGPT is at least 2 years, and domestic technology manufacturers should increase investment in the research and development of underlying technologies, shorten the generation gap, and show more responsibility and action.
Photo: Dr. Zhang Jun, Chairman of CEIBS Capital and former Vice President of Huawei
Two years ago, it was predicted that there was a bubble in the metaverse
ChatGPT brings new opportunities
In stark contrast to the popularity of ChatGPT, the metaverse, which was previously very hot, seems to be a little "lonely". In the past year, many technology and entertainment companies have begun to retreat from the metaverse and have reduced their investment in metaverse projects. Since the beginning of this year, Microsoft has dissolved the core department of the newly established industrial metaverse; Disney fired the team developing the metaverse strategy; Kuaishou's metaverse project leader left, and metaverse-related businesses were also suspended...
"As early as two years ago, I predicted that there was bubbles and hype in the metaverse." Zhang Jun said that from the market, the meta-universe has indeed ebbed, and the landmark event is that Meta (formerly known as Facebook) laid off more than 10,000 people.
Zhang Jun believes that the metaverse itself is not a technology, but an application scenario of technology integration, such as VR, AR, MR, digital human, digital twin, and even ChatGPT, which is only the technology to realize the meta-universe. At present, it is still the early bubble stage of the meta-universe, and there are too few landing application scenarios, and how the meta-universe will land depends on the development of technology.
"Indeed, after AIGC, especially ChatGPT, came out, artificial intelligence technology has made a milestone leap, pushing the concept of the metaverse one step forward from the perspective of productivity. But from now on, the metaverse is still a relatively vague concept, and it has not yet formed an industry. ”
He said that for the metaverse, it is necessary to take a long-term view. At present, more new technologies like ChatGPT are needed to realize the grand vision of the metaverse.
Domestic large model tracks are hot
It is recommended that enterprises do what they can and should not blindly follow the trend
With the explosion of ChatGPT, exploring large models and ChatGPT-like products has become a hot topic in the field of AI. Major manufacturers have set their sights on the big model track, such as Baidu, Huawei, Ali, etc. have announced large model plans. In addition to large factories, the presence of small and medium-sized enterprises is not uncommon. According to statistics, up to now, more than 40 domestic companies and institutions have entered the field to announce their entry into the large-model track. However, Zhang Jun believes that to create a large model, enterprises should do what they can and should not blindly follow the trend.
"These small and medium-sized companies, if they want to start a business to do ChatGPT and make a big model, I think they should be cautious. Compared with large companies, small and medium-sized companies have a big gap in terms of manpower, financial resources, and material resources, and now everyone rushes up, and even some people who don't understand at all, I think enthusiasm is good, but interlacing is like a mountain, not everyone can do it. ”
He ridiculed, "Back then, Dong Mingzhu said that he would invest 10 billion yuan a year to make chips, and I told her at that time that at least a '0' would be added later, and at least 100 billion would be spent, otherwise don't do it." ”
Recently, OPPO announced that it would stop developing mobile phone chips and dissolve the chip team, which attracted attention in the industry. Zhang Jun said, "This OPPO broken arm, in fact, there is a lesson from the past, that is, Xiaomi core manufacturing also failed. If OPPO learns the lesson of Xiaomi making mobile phones first and then mobile phone chips, he will not burn the 50 billion. ”
So, what kind of enterprise is suitable for making a big model? In Zhang Jun's view, first-line enterprises can consider doing it. He said that in the United States, it is not some startups that are making large models, and it can be seen from OpenAI's loss of hundreds of millions of dollars last year that large models are very money-burning, depending on whether the company's human, financial and material resources are strong enough.
He believes that it is not necessary for everyone to do AIGC, or conversational robots, because this is only a very small branch of artificial intelligence, but ChatGPT is indeed an iPhone-level phenomenon of artificial intelligence, which will bring a leap in productivity and thus force the progress of production relations.
"As an entrepreneur, you still have to do what you can and be down-to-earth, rather than following the trend, doing AIGC in order to do AIGC, and doing ChatGPT in order to do ChatGPT." He suggested that domestic small and medium-sized innovative enterprises need to find a balance between technological breakthroughs and survival and profitability, such as solving the "stuck neck" problem of the underlying technology and inventing breakthrough processes to achieve domestic substitution, or overcome industry shortcomings.
Large domestic models are two years behind foreign ones
Call on science and technology manufacturers to pay attention to the research and development of the underlying technology
For the domestic science and technology manufacturers that have entered the large-model track, Zhang Jun's suggestion is: increase investment in the research and development of the underlying technology.
In Zhang Jun's view, although the mainland has made great progress in the field of artificial intelligence, and has been on par with the United States in some tracks, in the field of large models such as dialogue robots, the gap between us and foreign countries is at least 2 years. "ChatGPT iterates every few months, and two years is actually a long time."
He said that the emergence of ChatGPT as an application scenario requires the support of higher computing power, and the underlying core technology is the software and hardware platform. The core of the hardware is the GPU chip, and the software is the embedded operating system, database, middleware, etc. In terms of computing power, the current domestic platform cannot support us to do ChatGPT. The domestic large model is two years behind the foreign ChatGPT, lagging behind in the "lack of core and less soul", and there is a big gap with foreign countries in terms of manpower, financial resources, material resources and other aspects.
Technically, some of the world's top GPU chip products will cut off supply to domestic enterprises, and even provide "castrated version" chips; In terms of funds, training a ChatGPT costs tens of millions of dollars, and ordinary companies do not have such financial resources; In terms of manpower, a basic research team with large models is needed.
"We started late because our Internet giants did not pay attention to the real underlying core technology, but focused on sharing bicycles, OTO, and grocery shopping. Internet giants should not only worry about the traffic of those bundles of cabbage, but should have more responsibility, more pursuit, and more achievements in scientific and technological innovation. He believes that few people in China can stand at the height of Steve Jobs and Musk to think about science and technology, and even the future of mankind. Therefore, the generation difference between domestic large models and foreign ChatGPT in the past two years is not only technical, but also conceptual and patternal.
On CEIBS Capital's Vision:
Investing in "the next Huawei"
Zhang Jun is a postdoctoral fellow at Shanghai Jiao Tong University, who has worked in Silicon Valley, Shanghai, Shenzhen and other places in the United States, and has been engaged in the communication industry for about 40 years. He was a vice president at Huawei, where he was in charge of Huawei's software company. He said, "Huawei's operating system Hongmeng was proposed by me more than ten years ago, but unfortunately no one at Huawei understood my words more than ten years ago. Zhang Jun said that Ren Zhengfei also apologized for not letting him be a Hongmeng.
After retiring from Huawei, Zhang Jun founded CEIBS Capital in 2015 and currently serves as Chairman of CEIBS Capital. According to the data, CEIBS Capital focuses on strategic investment in hard technology fields such as artificial intelligence, semiconductors/integrated circuits, Internet of Things, and 5G.
Zhang Jun said that as a new player in the VC industry, CEIBS Capital has just registered for the record, but has gathered 18 industry experts, all of whom are experts defined by Huawei. "We are not academicians, professors, doctoral supervisors, researchers, scientists, we are engineering businessmen, Jobs, Ren Zhengfei, Musk are all engineering businessmen, Jobs is the most typical is that the iPhone can be made, do the best in the world, but also sell it the best, this is called engineering businessman, this is Huawei's definition of experts." Through 18 engineering businessmen, CEIBS Capital not only provides funds to the invested enterprises, but also empowers and adds value to them in post-investment management.
He said that the criteria for capital investment in China and Europe are "the top three in the world and the first in China", such as the recent investment in high-end motor drives, ultra-short throw lenses/optical machines, proportional valves for medical ventilators/lithography machines, etc., which are all benchmarked against international firsts and domestic leading projects. The criterion for judging investment mainly depends on whether the technology is ahead, not the market share.
Zhang Jun said that CEIBS Capital's vision is to invest in the next Huawei. "Because to realize the Chinese dream, China needs at least 10 to 20 companies like Huawei, and China now has only one Huawei, so there is still a long way to go, but the road is long and there will be glory."
Written by: Ao Yi News, Lu Ruoqing, Lin Sisi