The first anniversary of ChatGPT’s “surge”: Can domestic performance really catch up with foreign countries? Three days later, it will be the "one-year anniversary" of OpenAI announcing the official launch of ChatGPT. In the past year, the number of large and small models in China has already exceeded 100. Domestic large models represented by Wen Xinyiyan, iFlytek Spark and Tongyi Qianwen have successively announced that their performance has surpassed GPT3.5, and the issue of commercialization of large models is also constantly being mentioned...
After the storm, many large model companies in China announced their dissolution, and problems such as "sheathing" foreign open source projects by well-known large model projects occurred frequently.
The development of domestic large models is also encountering a new "bubble". Looking back on the year since ChatGPT was released, what keywords would you use to summarize?
Awakening and Fury
The number exceeds 100 and applications emerge.
One year ago today, with the release of ChatGPT, the emergence of intelligence and natural language interaction capabilities it brought quickly set off a wave of AI 2.0 catching up around the world.
Domestically, while a large number of C-end consumer groups have jumped over the wall to register and experience ChatGPT, major Internet companies represented by Baidu, Alibaba, iFlytek, and 360, as well as a few start-ups such as Lanzhou Technology, have quickly launched their own large-model products.
On March 16 this year, Baidu's large language model product "Wenxin Yiyan" was first released, and the first batch of 650 companies announced their access to the Wenxin Yiyan ecosystem. Subsequently, on March 29, 2023, version 1.0 of the 360 Intelligent Brain model was released. On April 11 this year, Tongyi Qianwen, a subsidiary of Alibaba, appeared at the Release Cloud Summit. Alibaba announced that all future products will be connected to the "Tongyi Qianwen" large model. As of May 6, iFlytek's large-scale cognitive model of iFlytek will be released...
Behind the rapid follow-up, the emergence of ChatGPT was not instantaneous. The underlying AI large model technology that supports its development has been developed for many years. Domestic first-tier Internet companies and cutting-edge start-ups have already laid out their plans, which has become the key for domestic manufacturers to quickly follow up this time.
According to previous statistics from Sina Technology, as of July 31, more than 100 large model products have been launched in China. Among them, there are as many as 26 companies targeting general large models and "general + industry" companies; in addition, there are 7 "basic + industry large models"; there are as many as 67 large models in finance, medical, biology, transportation, data analysis and other industries.
At the same time, more and more AI native applications based on large model technical capabilities are rapidly emerging. According to Baidu's announcement on October 17, one month after the company's Wenxinyiyan plug-in ecological platform "Lingjing Matrix" was released, the number of developers applying for Lingjing Matrix has exceeded 27,000, covering multiple fields such as law, workplace, and learning.
In communication with Sina Technology, a planner who is organizing an AI entrepreneurship competition said based on his own practice: "Just a few days after the conference started recruiting, hundreds of AI application companies signed up to participate, which is really surprising."
Behind the continuous emergence of AI native applications, large model technology has changed the development paradigm, allowing more non-technical personnel to begin to have the ability to quickly develop AI native applications, which has become an important reason for the rapid emergence of AI native applications.
catching up and surpassing
Domestic large models claim to catch up with GPT3.5
With the rapid emergence of domestic large model manufacturers and AI native applications, it is also becoming an established fact that high-quality domestic large models are rapidly catching up with GPT3.5.
In July this year, Baidu announced that the new version of Wenxin Yiyan 3.5 had surpassed ChatGPT 3.5; by October 17 this year, Wenxin Big Model 4.0 was released, and Baidu once again shouted the slogan that its comprehensive capabilities are not inferior to GPT-4. In addition, on October 24, with the release of Spark Cognitive Large Model V3.0, the company announced that this version of the large model has surpassed ChatGPT (GPT3.5). On October 31, with the launch of Tongyi Qianwen App, Alibaba also announced that Tongyi Qianwen surpassed GPT-3.5 and was partially comparable to GPT4.0.
Is the gap between domestic companies and global leading companies such as OpenAI really narrowing?
In this regard, Jiang Guanjun, head of research and development of Alibaba's Intelligent Information Business Group, told Sina Technology and other media, "In the past half year in China, the level of Chinese large-scale models has developed rapidly, and everyone can have an intuitive feeling. In the first half of the year, there were high levels of errors, inaccuracies, and unsmoothness in various apps, but now, it is relatively good."
In Jiang Guanjun’s view, the overall capability level or general intelligence technology level of domestic large-scale head models is definitely better than GTP3.5. This is certain. "The GTP3.5 online service model is not large, and I believe this news. When the domestic large model reaches more than ten B, everyone will find that the Chinese ability and English ability are very close to GTP." Jiang Guanjun said.
Bankruptcy and Settlement
A new AI bubble is forming
In addition to the positive side, in the past year, issues such as the dissolution and bankruptcy of large model and generative AI companies, and the "shelling" of well-known domestic large model projects into foreign open sources have also frequently attracted attention from the outside world. The development of domestic large models is also encountering a new "bubble".
In June this year, "Light Years Away", a large model company founded by Wang Huiwen about two months ago, officially announced that it would be acquired by Meituan as Wang Huiwen fell ill. A month later, it was revealed that the first-generation AIGC Unicorn Film Technology, which had received an investment of 1.36 billion yuan in the D round, had cut off the social security and provident fund payments of its employees, and many of the company's employees resigned and left. Abroad, AIGC star unicorn Jasper.AI, which has received two rounds of financing of US$85 million and US$125 million in 2021 and 2022 respectively, with a corporate valuation of US$1.5 billion, has also officially announced layoffs and downsizing.
Recently, Jia Yangqing, the former vice president of Alibaba and a well-known AI framework expert, broke the news in his WeChat Moments that a large domestic factory has built a large model called LLaMA, which has triggered industry doubts about Kai-fu Lee's "Zero One Thousand Things" large model built into LLaMA, which is also attracting more and more widespread attention.
In this regard, Shen Meng, executive director of Chanson Capital, said in a communication with Sina Technology: "This means that the bubble of domestic large-scale models has arisen. Although domestic AI entrepreneurship is currently very popular, and various companies must call it AI, it is actually not worth a try. Even many small and medium-sized investors know in their hearts, 'I know you don't have AI, and you also know that I know you don't have AI', but everyone still continues to spread the word and blow the bubble."
"We hope that we can go deep into the bottom-level enterprises and try to make as many bottom-level products as possible, instead of just taking other people's bottom-level code and packaging it as our own. This is the most unacceptable thing. If we keep floating on top and keep taking other people's things, then the foundation of our technology building will always be fragile and it will be difficult to support the growth of the entire AI technology tree." Shen Meng said.
In Shen Meng's view, the reason why AI has become a very hot concept in China is inseparable from the very high expectations of all walks of life in the country for AI. I hope that domestic AI technology companies can put some good products and concepts into society and contribute to the healthy development of AI.
"Many entrepreneurs think that they only need to cover their projects with AI to get investment. This is too optimistic. They understand neither AI nor the market nor investment. In the past, the market environment in which you could get a lot of money just by copying a PPT has passed. We should not do too much bubbly work." Shen Meng said.
Infighting and restructuring
Governance and value alignment are urgent
This week is about to end, and the bizarre "coup" infighting among OpenAI management continues to become the focus of the world. Behind this, the board of directors, represented by OpenAI chief scientist Ilya Sutskever, who mainly pursues AI safety and explainability, and the group represented by Sam Altman, who is eager to accelerate the commercialization of AI, quickly moved into confrontation due to different ideas.
It's just a pity that in the end, the group represented by Ilya Sutskever, who mainly pursues AI safety and explainability, seemed to finally reach a compromise in the drastic changes staged by OpenAI management. Forced by nearly 700 OpenAI employees to resign, OpenAI announced that Sam Altman will return to OpenAI as CEO, and the company's board of directors will also be reorganized.
"The repeated and continuous changes in OpenAI's management have exposed serious loopholes in OpenAI's complicated governance structure. I feel very disappointed. It feels like a group of highly intelligent people doing an unprepared and unorganized thing. Can a company with such a loose and unorganized organizational structure really be able to carry people's expectations for general artificial intelligence technology?" In communicating with Sina Technology, Pharos, the manager of Capricorn Business Review, expressed his disappointment.
As the "coup" of OpenAI's management unfolds, issues such as AI governance and value alignment are also attracting more and more attention. In Shen Meng's view, "A set of global standards similar to the 'Three Laws of Robotics' should be formed as soon as possible globally to form constraints and governance on artificial intelligence technology at a certain level to avoid the abuse and misuse of AI technology."
Conclusion
Embrace the AI big model, but you need to put "reins" on it
From the perspective of technology and industrial implementation, although the country has made certain breakthroughs in catching up with ChatGPT, objectively speaking, "the country is still lagging behind OpenAI and others." Under such circumstances, with the return of Sam Altman and the continued advancement of OpenAI commercialization, if the country blindly emphasizes governance and security and ignores the development of large AI models, it will eventually fall behind.
For each individual, they should think more about how to make good use of AI, arm themselves with new AI technology, and embrace and control AI well. For companies engaged in the R&D of large models, they should also define their own moral bottom line in the process of promoting commercialization and refrain from doing things that harm their users and customers.
Text丨Sina Technology Zhou Wenmeng