At the shareholders' meeting in mid-May, when asked why Tencent was slightly low-key under the ChatGPT trend, Ma Huateng replied: Don't worry.
Internet companies have a lot of accumulation, are doing, we are also immersed in research and development, but we are not in a hurry to finish early, and take out the semi-finished products to show.
Pony Ma believes that the key to a large model is to do a solid job in the underlying algorithms, computing power and data, and more importantly, the scene landing.
I believe that we have a lot of scenarios that can be landed, and we are still doing some thinking.
But just a month later, Tencent figured it out.
On June 19, Tencent Cloud held the Industry Big Model and Intelligent Application Technology Summit at the National Science and Technology Communication Center, announcing the R&D progress of Tencent Cloud's industry big model for the first time.
To be precise, it is an industry large model service solution designed by Tencent: unlike Baidu, iFLYTEK or OpenAI, Tencent did not release the rumored mixed element large model or a certain GPT at the conference, but repeatedly emphasized that its own large model is designed for "specific applications designed for specific scenarios" around the needs of the B-side.
From the timeline, Tencent is one of the latest technology giants in China to announce the progress of large models, and in the second quarter of 2023, when friends concentrated on releasing large models, Tencent's stock price fell by 11.12% in three months.
This meeting stacking application scenario finally has an explanation with the market and investors. Pony horses can say that they are not in a hurry, but within the company, it is actually impossible not to be in a hurry.
Tencent, made a consumer product
"AGI also needs scenario implementation, targeted data screening, and optimization algorithms."
Previously, when AI Blue Media Exchange (ID: lanmeih001) had a conversation with an AIGC founder, the other party talked about the application of ChatGPT-type general large models, and judged that the next step of AGI (general artificial intelligence) is to find specific applications and optimize them.
Tencent's large-model strategy, on the other hand, has chosen a different route: bypassing generic large-model products similar to friendly companies and directly customizing them according to the needs of the industry.
Let's not speculate about Tencent's progress in the general big model, at least Tang Daosheng, senior executive vice president of the group and CEO of the cloud and smart industry business group, said that the general big model is actually not so necessary:
Although everyone has high expectations for the general large model, it is not necessarily the optimal solution to meet the needs of industry scenarios.
He believes that what companies need is to actually solve a problem in the actual scenario, not solve 70%-80% of the problem in 100 scenarios.
The route chosen by Tencent is to find customers directly: throw out the concept of "big model selection store", that is, buy it and use it immediately, and make it into consumer goods.
The official explanation is this: it's like walking into a super clothing store and there are "recommended models" at the door... Guests who come in can also DIY customized "exclusive models" according to their own preferences.
The scene proposed by Ma Huateng a month ago landed and was finally verified.
Tencent's message is that customized solutions are better to use and solve problems than to advocate a generic model. Tang Daosheng said that in the specific implementation, "models, data and computing power are three points that everyone needs to pay special attention to." ”
Models are closely related to data. Since ChatGPT was first released, insufficient information accuracy has always been a problem that the industry has not solved, and Zhang Guan Li Dai has occurred from time to time. Tencent's solution is to subdivide the database and use targeted vertical industry or enterprise data to make the output of the model more detailed and accurate, easy to manage and correct, reduce the volume and improve efficiency, which also means the reduction of computing power costs.
Computing power is the hardware foundation of the era of large models, and "high-performance, high elasticity and high stability of computing power are essential for the training and use of large models".
The combination of databases, cloud services and high-performance computing clusters in vertical industries is convenient for customer applications and has become the selling point of Goose Factory's large model.
That's certainly true – because friends are doing the same. Tencent can bypass the release of general large models, but it cannot bypass the competition in application fields.
Fight fast and grab customers
At first, we thought that this was an opportunity for the Internet that would not happen in a decade, but the more we thought about it, the more we felt that this was an opportunity similar to the industrial revolution that invented electricity in hundreds of years, so we felt that (AI) was very important.
Similar to the same industry insiders, externally, Ma Huateng regards the window brought by the big model as another singularity of technology and even civilization, and internally, it is also an opportunity for the company to expand its business and even change the industry pattern.
Happiness must be in spring, and the wind should be early. But obviously, in the debut of the big model, Tencent has been slow.
The next competition is on the application side - finding someone to pay for this matter must not be delayed any longer.
Regarding applications, the industry's solutions are actually similar in thinking, but the specific solutions are slightly different: Microsoft has integrated ChatGPT into search and office, and companies such as Baidu and iFLYTEK have also successively integrated their own large models into writing, translation, marketing... Bringing AI into the field that it is good at and needs to be incremented has become a tacit understanding in the industry.
Tencent, chose to cover the large model to enterprise-level applications and cloud services, and Tang Daosheng even specifically "suggested" enterprise managers to use cloud vendor tools to build integrated model services in his speech: "This is relatively efficient, saving the cost and time of training, operation and maintenance." ”
Previously, at the earnings conference, Tencent expressed its hope to increase the revenue margin of the cloud business, reduce loss-making projects, and further pursue "high-quality" growth. Relevant reports show that Tencent Cloud's domestic public cloud IaaS + PaaS market share in the second half of 2022 has decreased from 11.1% to 9.9%, and its ranking has also fallen from second to fourth.
By 2023, a large number of friends in the technology and Internet industries are relying on the concept of large models and technologies to seek increments. Although Ma Huateng believes that "there are many companies that are too anxious, it feels that in order to boost the stock price, we have always not been in this style", although the Martin Lau said that "we are not in a hurry to go to the stage immediately to grab a seat, the seat is already there, the scene, users, and services are already there", but as soon as possible, it is always better.
After all, as a leading manufacturer in many fields such as gaming, social networking, and work, Tencent's delay in any step in the wind will not be good for market sentiment and business development.
For the announcement of the progress of this large model, the market reaction was actually mediocre. At the close of trading on June 19, Tencent Holdings fell 1.6% in a single day, and as of the 21st, the decline had reached 6.23% in three days.
Some investors believe that Tencent, which has a large number of social and office scenes, is a pragmatic choice. But some users said, "After the concept, it's time to look at the order." ”
Source|AI Blue Media Exchange Author|Yiqi