The Tencent Mixed Element Model is coming

Author | HUANG Yu

Edit | Zhang Xiaoling

In May, Ma Huateng, who said that he was not in a hurry to take out the "semi-finished product" of the general large model, was still in a hurry.

After the first batch of domestic AI large models were opened to the public, Tencent, which was slow for a beat, finally made a move.

At the 2023 Tencent Global Digital Ecosystem Conference held on September 7, Tencent Cloud unveiled the mystery of its self-developed universal large model - Hybrid Element.

At this time, it is half a year late from Baidu Wenxin, which was first released in China, and the mixed element large model cannot be called a fully mature large model. Of course, no big model company has the confidence to give an affirmative answer.

Jiang Jie, vice president of Tencent Group, also said frankly: "The version of the hybrid model we launched now is only usable and practical. ”

This release is only the beginning, the hybrid model is still only open to the industry through Tencent Cloud, and has not yet opened the ToC business to the outside world.

This release, in Tencent's internal view, is the most important move of its entire To B business this year, which means that Tencent's transformation of layout for many years, with the blessing of AI, has come to a new node. Obviously, Tencent does not have the mood to be in the mood, and the accuracy of the ability of the mixed element model and the hardcore of commercial value are what it pays the most attention to.

Pragmatist

The belated Tencent Mixed Element model emphasizes practicality, calling itself a practical-grade model that "comes from practice and goes to practice".

This is the consciousness of latecomer Tencent, and it is also the development trend after the big model boom fades.

According to CCID Consulting, as of July 2023, a total of 130 large models have been launched in China. Under the "100-model war", only large models with real practical value can stand out and finally be accepted by people.

Application scenarios are the key factor to victory, which is also the advantage of Internet giants like Tencent.

At present, the hybrid model has been connected to more than 50 businesses of Tencent and has achieved preliminary results, including Tencent Cloud, Tencent Advertising, Tencent Games, Tencent Fintech, Tencent Meeting, Tencent Docs, WeChat Souyi Search, QQ Browser and other businesses and products.

It is understood that the hybrid model will be used as the foundation of Tencent Cloud MaaS services, and customers can not only call the hybrid element directly through APIs, but also use the hybrid element as the base model to build exclusive applications for different industry scenarios.

Tencent also launched the "Tencent Cloud MaaS Service", which not only provides mixed element large models, but also shelves more than 20 kinds of the latest and most popular open source general large models in more than 20 industries, as well as industry large models covering more than 20 fields such as finance, cultural tourism, retail, government affairs, medical care, and education.

It is not difficult to find that most technology companies will open up their Chat capabilities for long-term external verification before opening the general large model to the industry, and at the same time accumulate user data to help the iteration of the large model. For example, OpenAI recently launched the ChatGPT Enterprise Edition after launching ChatGPT to ordinary consumers for some time.

Tencent's hybrid model seems to have taken a completely different path.

Jiang Jie believes that there are many ways to test large language models, one of which is to verify through Chat, which is a relatively good body feeling. Open AI is a startup, but Tencent has a large number of users, and it may be more effective to choose to combine Tencent's massive applications to do some tests.

Therefore, it can be seen that the Tencent hybrid model, which chose to skip the external verification stage, also emphasized its application in the industry when it debuted.

As Jiang Jie said, Tencent's hybrid model must first serve Tencent itself, and then open up to the outside world through Tencent Cloud to serve customers and ecological partners.

This is the model chosen by Tencent to make a general large model. As for the C-end, Tencent's hybrid model has no plans to open for the time being. In Tencent's planning, the ability to open the hybrid model to the C-end is to be realized through WeChat and QQ. However, Tencent has not yet obtained the approval to open to the C-end.

In Jiang Jie's view, as long as Tencent's general large model has sufficient capabilities, whether it can be opened to the C-end is only a process of pressing the "switch".

In the Internet era, Tencent has always been reluctant to be left behind in testing new products. Now entering the AI era, Tencent has become the most "cautious" of the three giants of BAT.

However, every second counts in the age of AI. Just when Tencent's mixed element model is still being tempered internally, the industry has ushered in a critical moment.

On August 31, a number of AI large model manufacturers, including Baidu, SenseTime, Zhipu, MiniMax, etc., officially announced the launch of large model products for the whole society. Not long after, 360 Smart Brain and iFLYTEK's "iFLYTEK Spark Cognitive Big Model" were also opened one after another.

Tencent and Ma Huateng also had to make a move.

Feelings of anxiety

Whether it is mixed yuan or Wen Xin's words, the Internet giants have bet on the big model, behind it is just the main business peak, the anxiety of how to grow in the future, and the fear of being abandoned by the times, which is not fundamentally different from Microsoft's investment in Open AI.

Pinduoduo and Douyin e-commerce led Alibaba's revenue from Chinese retail commerce in fiscal 2023 to decline by 2%. Account management revenue decreased 8% year-over-year.

Tencent's revenue and net profit fell in 2022, and the revenue of its game business, an established cash cow, also fell by 2%, surpassed by the fintech and enterprise services business for the first time.

Whether it is the older generation of BAT, or the later Byte and Kuaishou, Chinese Internet companies seeking a second growth curve on the B side seem to be an inescapable fate.

But the business is getting harder and harder to do. The majority of the IaaS (Infrastructure as a Service) field is seriously homogenized, the price war has been defeated, and state-owned clouds and operator clouds have risen rapidly and attacked cities.

Last year, the market share of the bellwether Alibaba Cloud shrank from 36.7% to 31.9%, and the share of Tencent Cloud decreased from 11.1% to 9.9%, and the ranking fell from second to fourth.

How to find new growth points and how to sell high-margin products is a question that Tencent Cloud must answer.

Reshaping To B business with AI is the common answer given by Internet giants. Industry big models have also emerged, usually between the basic layer of the general large model and the more specific AI application service layer.

Before Tencent Mix, Alibaba Cloud had already demonstrated the "Tongyi Qianwen", whose industry models were divided into more than ten fields such as e-commerce, healthcare, entertainment, design, and finance.

Baidu's "Wen Xin Yiyan" provides external services through Baidu Intelligent Cloud to help enterprises build their own models and applications, and has released more than ten industry models in the fields of electric power, finance, and media.

Compared with the grandeur and abstraction of the previous wave of "meta-universe" hot spots, in this round of AI wave, ChatGPT has personally demonstrated the practice of quantitative change leading to qualitative change. This has given all walks of life greater enthusiasm for landing, and also brought cloud computing the imagination to escape involution - with AI service capabilities, from Internet customers with exhausted dividends to a broader real economy, into the deep water area of industrial digitalization.

However, if there were no Internet giants slowing down in the growth rate of their main business and falling into the dilemma of involution in cloud services, the "G wind" blowing from the other side of the ocean may not be so strong.

They have applied large models to all walks of life, and it is undoubtedly an echo of this pressure.

C-end traffic has peaked, and the ToB business, which is the "second growth engine", has no way back.

In this sense, the "big model release index" of Internet giants may be a "growth anxiety index".

It's a new era, and everyone is on the same page again.

This article is from Wall Street News, welcome to download the APP to see more