With the opening of Pandora's box of AI large model, the cloud market GameChanger has arrived. For Alibaba Cloud, the former big brother, is there still a chance to reshape the industry rules from a "leader" to a "follower"?
This may be the most important "Stud" concerning the future since Zhang Yong took full helm of Alibaba for 7 years.
"All Alibaba products will be connected to Tongyi Qianwen for a comprehensive transformation." A week ago at the National Convention Center in Beijing, at the Alibaba Cloud Summit, Zhang Yong, who made his public debut as the CEO of Alibaba Cloud for the first time, stood in the center of the stage and unveiled the The veil of the large model of "Thousand Questions of Tongyi".
This is the first work to be verified after Zhang Yong took charge of Alibaba Cloud. According to Zhang Yong’s announcement, 27 APP products under the Ali Group, including Tmall, Dingding, AutoNavi Maps, Taobao, Youku, and Hema, will all be connected to the large model of “Tongyi Qianwen” for a comprehensive transformation.
Aggregating all Ali products under the same technical framework is very similar to the famous "big, medium and small front desk" organizational strategic path proposed by Zhang Yonggang when he was at the helm of Ali in 2015. Success, as early as 2020, Zhang Yong began to reflect on himself, and in this year's unprecedented organizational transformation, he completely split into "1+6+N".
If it is said that the launch of "Big Middle Platform" was an innovator and leader, but in today's GPT large-scale model arms race, Alibaba Cloud has fallen behind a lot, and has even become an imitator and follower.
Not only is the large model released later than foreign OpenAI, Microsoft, Google, Meta, as well as domestic Baidu, Huawei, and SenseTime, but even Zhou Jingren, Chief Technology Officer (CTO) of Alibaba Cloud Intelligence, proposed the "integration of cloud and intelligence" at the summit. Strategy - the strategic direction emphasized by Alibaba Cloud Intelligence Group. In fact, Baidu officially released the "cloud-intelligence integration" strategy as early as 2020, and Alibaba Cloud was three years later than Alibaba Cloud.
Why did Alibaba Cloud propose the "integration of cloud and intelligence" strategy at this time? Is it "copying homework" or "industry consensus"? What impact will it bring to the large-scale model arena and the cloud market structure?
01 | Cloud-Intelligence Integration——Alibaba Cloud "Copying Concepts"?
"In the era of intelligence, everyone is on the same starting line." Facing the venue crowded with thousands of people, Zhang Yong said that in addition to all products of Alibaba that are worthy of "transformation" with large models, Alibaba Cloud will open Tongyi Qianwen Capability, access to B-end enterprises to create exclusive large models.
Through the speeches of Alibaba Cloud executives such as Zhang Yong, we can get a glimpse of Alibaba Cloud's great ambitions: through the "All in" of the big model, we can bet on the future of Alibaba Cloud and even the Alibaba Group.
As we all know, Ali, which had a first-mover advantage in the early years, relied on the explosion of e-commerce to stand at the pinnacle of China's Internet. Its aggressiveness, methodology, organizational ability, political commissar system, mission vision, values, etc. have influenced countless entrepreneurs and entrepreneurs.
This also includes the strategy of "large, medium and small front desks" established by Zhang Yong after he became the CEO of Ali Group in 2015. The purpose is to improve the overall efficiency of Ali Group, minimize the repetitive behavior of building wheels, and reuse existing capabilities—— Although "Dazhongtai" was not created by Ali, but came from Ma Yun's experience after visiting a game company in Finland, "Dazhongtai" became popular because of Ali, and is still learned and imitated by many companies.
In recent years, Ali has been weak in innovation and has not been able to keep up with the "evolutionary iteration" of the market and the general environment. It has frequently lost in various battlefields, and its market value has shrunk by nearly 70% from its peak.
For example, in the field of e-commerce, JD.com, Pinduoduo, and Douyin e-commerce have seriously shaken their fundamentals; in the payment market, Alipay’s share has long been dropped by WeChat Pay; in the field of local life, from Ele.me to Fliggy, they are powerless to stop Meituan’s attack; In the field of cloud computing, the growth rate of Alibaba Cloud is declining, executives are in turmoil, and Tencent Cloud, Baidu Smart Cloud, and Huawei Cloud are under great pressure; in the field of entertainment, the development is even more bleak, and acquisition failures abound.
Under various difficulties, on March 28 this year, Ali ushered in the largest adjustment of the company's organizational structure in history. For Ali, which adjusts its organizational structure every year, it is easy to adjust to "1+6+N", but the most difficult thing is how to find a new development path.
When the AI model represented by ChatGPT is regarded as the cornerstone of the next-generation Internet, and is being competed by all technology giants in the world, Ali will naturally not miss it, just as Zhang Yong made the prediction at the summit: the current technology field is It is the juncture of forming a major historical opportunity for the next 10 or 20 years.
However, compared with the large-scale model products launched by OpenAI, Microsoft, Google, and even Baidu, "Tongyi Qianwen" is regarded by many people as a "Great Leap Forward" in the outside world.
According to Tongyi Qianwen’s own introduction, this product was developed by Alibaba’s Dharma Institute and was born in November 2022. In February of this year, a senior technical expert in Ali broke the news that Dharma Academy was developing the Ali version of the chat robot ChatGPT, and Ali later confirmed that it was in internal testing. On April 7, Alibaba Cloud suddenly announced that the large-scale model Tongyi Qianwen was invited to test, and the content has since been deleted by the publisher. On April 11, Tongyi Qianwen appeared in Zhang Yong's Aliyun debut with high specifications.
Strictly speaking, there are not many innovations in the technical route and product form of "Tongyi Qianwen". In terms of technical route, the products that have been launched so far include ChatGPT-like mode and multi-modal hybrid mode. Tongyi Qianwen is similar to Wenxin Yiyan released by Baidu in March (digression: there are Wenxin Diaolong in famous ancient literary theory works. , Baihu Tongyi), is a large-scale pre-trained language model similar to ChatGPT, which has extensive knowledge reserves and universality. Scenario requirements.
Even though Zhang Yong proposed to help enterprises create a general-purpose platform for exclusive GPT, in fact, there are cloud enterprises that are similar to "open source" at the forefront. At the end of March this year, Baidu Smart Cloud officially launched the "Wenxin Qianfan" large-scale model platform to provide customers with enterprise-level large-scale language model services.
It is worth noting that Alibaba Cloud also clearly stated for the first time that "integration of cloud and intelligence" is its future strategy. Including Zhou Jingren, chief technology officer of Alibaba Cloud Intelligence, who gave a speech on "Integration of cloud and intelligence, making intelligent innovation within reach", and Cai Yinghua, chief commercial officer of Alibaba Cloud intelligence, also said that only the integration of cloud and intelligence can make the overall industry intelligent.
The concept of "integration of cloud and intelligence" was first proposed by Baidu. In May 2020, Baidu Smart Cloud announced that CTO Wang Haifeng will lead the unification of cloud, AI and basic technology systems, and clearly put forward the "integration of cloud and intelligence" strategy. In the following two years, Baidu Smart Cloud completed the evolution of the cloud-intelligence integrated architecture from 1.0 to 3.0 at a frequency of one year.
The integration of "cloud + AI" is actually not uncommon in cloud computing companies at home and abroad. In addition to Baidu's "cloud-intelligence integration", Microsoft proposed "office+Teams+Azure", Amazon proposed "cloud, data, and intelligence trinity" in 2022, Tencent Cloud proposed "cloud-intelligence integration" last year, and JD.com called it "digital intelligence supply chain". Alibaba Cloud also had a "cloud nail integration" before - even if everyone has similar technical routes and concept meanings, they will avoid adopting the same concept.
In the past three years, in various occasions and financial reports, Baidu has been strongly bound to "integration of cloud and intelligence". This also puzzles the industry. Why does Alibaba Cloud re-propose the concept of "integration of cloud and intelligence" three years later? Is it "copying homework" or "industry consensus"?
02 | The logic still needs to be self-consistent — the gap between actual measurement and publicity is a bit big
In any case, from the perspective of industry insiders, Alibaba Cloud's follow-up means that the integration and integration of "integration of cloud and intelligence" has become an industry consensus, and it also heralds the arrival of a large industry opportunity for integration of cloud and intelligence.
Alibaba Cloud executives did not elaborate on the specific meaning of cloud-intelligence integration. According to Baidu's previous statement, "cloud" refers to cloud computing as the basis, enabling enterprises to access AI capabilities as conveniently and quickly as using "water, electricity and coal" on the cloud; "intelligence" refers to artificial intelligence. Full-stack technical capabilities such as vision, language, and knowledge can meet the needs of enterprises for AI technology in complex and changeable scenarios.
To put it simply, "cloud suitable for running AI" is superimposed with "AI that understands the scene" to jointly create the infrastructure of the intelligent era.
In the concept of "integration of cloud and intelligence", the core is intelligence. After years of layout and development, the cloud has become the infrastructure of all walks of life. Whether it can provide and deliver comprehensive, convenient, efficient, stable and reliable AI intelligent services has become the key to the competition of major technology companies.
The growth rate of revenue has declined, and Alibaba Cloud, which has gradually touched the ceiling in the government, enterprise and personal cloud markets, may see the same opportunity: relying on the differentiated approach of "integration of cloud and intelligence", Baidu has formed a "chip-framework-big model- Industry application" intelligent closed-loop path, helping Baidu find the second growth curve for the next ten years. It can be said that "cloud intelligence integration" is very important in the process of reshaping Baidu.
In addition, with the acceleration of China's economic and social digital transformation and the advent of the Internet of Everything era, "going deep into the industry and promoting the intelligent upgrading of the industry" has clearly become an important part of each cloud computing strategy.
This is even more inseparable from the "integration of cloud and intelligence" to carry large-scale model computing power and unlock AI productivity-in fact, whether it is an AI large model or other cloud and AI product matrices, although the concepts are different, they are actually all Launched based on the "Cloud Intelligence Integration" base. This is true for Wenxin Yiyan, Tongyi Qianwen, 360 Zhinao, SenseTime, and more other ChatGPT products.
Judging from the actual measurement of "Extreme Business", the answer given by Tongyi Qianwen is not ideal overall.
Tongyi Qianwen currently only provides text generation function, and cannot generate pictures, audio and video content for the time being. In terms of natural language processing, Chinese-English dialogue, etc., Tongyi Qianwen performs well, but it is not as good as other large models in terms of basic facts, time, calculation, reasoning, etc., which require thinking and logical understanding of dialogue. Compared with the current gap is not small.
"How to make fish-flavored pork shreds" is a stereotyped question that is widely mentioned in various models, and Tongyi Qianwen really answered it well. However, "Extreme Business" took the hot topic "Why did the Romance of the Three Kingdoms only focus on portraying the chaotic era in the early stage, while underplaying the final unification process?"
In contrast, Wen Xinyiyan pointed out that "the main plot of the novel is to describe the struggle for hegemony in the Three Kingdoms period, and the final unification process is just one of the small plots. This treatment makes the novel focus more on describing the history of the chaotic era The background and character of the characters, so as to better present the history and characters of the Three Kingdoms period."
On the issue of "How to view the "Generative Artificial Intelligence Service Management Measures (Draft for Comment)", the proposed ChatGPT provider should be responsible for the legality of the data source", another AI model gave affirmation, pointing out that it is the strengthening of artificial intelligence supervision. important move. But Tongyi Qianwen still "didn't learn how to answer".
Tongyi Qianwen may not know how to answer questions involving basic facts and timeline combing. For example, in the question "What negative events has Alibaba experienced in the past few years", Tongyi Qianwen's answer is not only full of mistakes, but even fabricated the facts and time of Alibaba in many places.
In the dialogue "Compared with Wenxin Yiyan, what are the advantages and disadvantages of Tongyi Qianwen", Tongyi Qianwen made another factual and low-level mistake: saying that both Tongyi Qianwen and Wenxin Yiyan were launched by Alibaba AI pre-trained model.
So many low-level mistakes are not only incompatible with Alibaba Cloud's first deployment of cloud computing in China and its market position, it is even more difficult to undertake the redoing of all Alibaba product access and the ambition to create exclusive large-scale models for B-end enterprises-after all, understanding the basic logic, Following the facts is the first priority. If the actual measurement results are unsatisfactory, it will not only affect the digital layout of B-end enterprises, but also bring bad experience to users such as DingTalk and Tmall Genie.
According to the layout of Alibaba Cloud, intelligent customer service is its key application scenario-it can help Ali break through the limitation of traditional e-commerce "people find goods" and iteratively upgrade to "goods find people". Moreover, in terms of wine travel, local life, etc., AI large models can also be used for precise marketing to meet the real needs of more consumers.
However, from the actual measurement, the performance of Tongyi Qianwen in terms of intelligent customer service and intelligent recommendation is still far away from commercialization. For example, "From Beijing, go to Chongqing Nantian Lake for a three-day trip to May Day" and "With a budget of 6,000 yuan, which range hood and gas stove are better to buy", the answers given by Tongyi Qianwen are practical for consumers. Very little reference.
It is worth mentioning that these questions were raised casually by "Extreme Business", not deliberately. At the previous press conference, Zhang Yong said that Tongyi Qianwen is "the world's first AI large model that breaks through 10 trillion parameters." In addition, it is reported that Ali has also taken the lead in building the largest AI model service community in China, "Magic Build". So, why is the logical ability of Tongyi Qianwen generative AI so weak?
03 | It's not easy to follow suit - what else is missing in Alibaba Cloud
This involves the essence of "integration of cloud and intelligence" - cloud must be better integrated with intelligence to give users a better experience and allow the industry to move towards intelligence in an all-round way.
The answer may be simple. Even though Alibaba Cloud has strong cloud computing and AI capabilities, the two have not been well integrated. In addition, the gaps in data parameters, the scale and quality of pre-training data, and data cleaning are also directly related to the answer results of different large models. Alibaba Cloud does not have an advantage in this regard.
Some insiders said that in the past, within some cloud vendors, AI and cloud were generally in two different systems. Selling resource-based products such as computing and storage has become the main business model, which has prevented AI from exerting its maximum value in the process of industrial digital transformation and intelligent upgrading.
For a long time, Ali's artificial intelligence and cloud computing have been in two independent systems - within Ali, the department for researching artificial intelligence was once in charge of the Ali Artificial Intelligence Laboratory (AI Labs) and Dharma Academy, which is the same as Alibaba Cloud. parallel sector.
AI Labs is a laboratory within the Ali system, researching voice, language, and machine vision, and later changed to the Tmall Genie Division to develop intelligent hardware. It was not until 2021 that the AI Labs team was merged into Alibaba Cloud Intelligence as a whole due to structural changes.
The Bodhidharma Academy, which develops Tongyi Qianwen, is an independent research institution outside of Ali's huge business system. The research scope includes data intelligence, human-computer natural interaction and other cutting-edge technologies. In the upgrade of the organizational structure in 2021, the "cloud and technology sector" in charge of former Alibaba Cloud President Zhang Jianfeng includes five lines: Alibaba Cloud, Bodhidharma Academy, DingTalk, Smart Internet (Tmall Genie), and Pingtouge. At that time, although the AI technology of Bodhidharma Institute was exported through Alibaba Cloud, it was still a parallel "brother department".
In addition to the parallelization of departments, what has had a greater impact on Alibaba Cloud is the turbulence of executives for several consecutive years. In May of last year, there were a series of executive changes between Dharma Academy and Alibaba Cloud. ) resigned, and Xiao Lihua (Dean of Alibaba Cloud Research Institute) left to start a business. Overnight, the star project that has been cultivated for many years since Wang Jian's time - Dharma Academy's Urban Brain Laboratory, almost all the core personnel were lost.
This made Alibaba Cloud and Bodhidharma "seriously hurt" and accelerated the "separation" of the two. According to a previous report by "Leifeng.com", Alibaba Cloud and Dharma Academy will divide their personnel at the end of 2021, and some products and technologies incubated by Dharma Academy will be set aside, while Dharma Academy will operate in a more independent form, taking on 10-digit costs. The challenge of revenue figures - the specific method is that Dharma Institute sells products through the Alibaba Cloud business line, and the two implement a dual-calculation model.
At the beginning of January this year, Alibaba Cloud's top management changed again. Zhang Yong personally replaced Zhang Jianfeng as the president of Alibaba Cloud Smart, and Zhou Jingren served as the CTO of Alibaba Cloud Smart. However, with the failure of the "big and middle platform", whether cloud and AI will be better integrated in the future has become a question.
After the "biggest in history" organizational structure change on March 28, DAMO Academy and Alibaba Cloud were both subordinated to the Yunzhi Group under the helm of Zhang Yong, but they still belonged to different business sectors and had a certain degree of independence in business. At the same time, executives are still missing, and Ali AI guru Jia Yangqing, and former leader of Ali M6 model Yang Hongxia have also left.
The core person in charge of the large model or AI has been in turmoil for two years. As a result, the launch of the Ali large model is late, and the actual performance of the product capabilities is more like "catch ducks to the shelves".
Some artificial intelligence practitioners revealed that a stable management team is the key to the continuation of product and technology routes. The person gave an example. In Baidu Smart Cloud, the combination led by Baidu CTO Wang Haifeng + Baidu Technology Committee Chairman Wu Hua + Baidu Group Vice President Wu Tian led the team. Only by maintaining stability can we have the continuous iteration of the Wenxin series.
Regrettably, for Alibaba's cloud and AI business, amid frequent organizational adjustments and continuous personnel turmoil, it never had Baidu's stable advantage.
This is also reflected in the large model. Ali is more than two years behind Baidu and Huawei in model iteration. In 2019, when Baidu and Huawei launched large-scale models, Ali started research and development. In 2021, other companies began to continuously iterate on the large model, and Ali released the language and multi-modal large model, which was merged into the Ali Tongyi large model in September last year.
The result is the difference in data quality and scale of large models. "Ali's large model is far behind the industry or domestic leading level in terms of existing data parameters, pre-training data quality, and data cleaning." The ICT industry service platform "Jiweiwang" reported.
Zhou Jingren, CTO of Alibaba Cloud, also admitted before that the research and development of large models often requires more than 100 billion parameters, which is far beyond the simple stacking of a single algorithm or GPU chips.
In fact, this gap is expected by the industry. Baidu has massive C-end search data, and Tencent also has more than one billion users of WeChat ecological data accumulation, but Ali has the most customer data of TOB merchants in the vertical field. From the perspective of general large-scale model training, TOB business data that is too commercialized is not suitable for mass addition.
Before Tongyi Qianwen was released, a senior expert from the Alibaba Cloud Research Institute admitted that in terms of parameters, Tongyi Qianwen was based on the PLUG model, with parameters ranging from 20 billion to 30 billion, while GPT-3 parameters were 175 billion. At the same time, its main It is biased towards text and does not involve images and videos; in terms of pre-training corpus data sets, it is inferior to OpenAI and Baidu in terms of quality and scale.
This may be one of the reasons why Alibaba Cloud "copy homework" and implanted the concept of "integration of cloud and intelligence". After all, as Zhou Jingren said, the current AI large-scale model competition is essentially a computing power competition. The competition is a systematic project that includes complex technologies such as underlying computing power, network, storage, big data, AI framework, AI model, etc. What is needed is the full-stack technical capability of AI cloud computing.
In any case, with the opening of the Pandora's box of AI large models, the cloud market GameChanger has arrived. A major reshuffle of the industry structure is inevitable. For Alibaba Cloud, the former big brother, how many opportunities are left to reshape the industry rules from a "leader" to a "follower"?