share_log

腾讯要开启“全民AI时代”

Tencent wants to start the “AI era for all”

wallstreetcn ·  May 30 06:42
The scene of the Yuanbao press conference. Liu Yuhong, Vice President of Tencent Cloud and Head of Tencent's Hybrid Model

Author | Huang Yu

Editor| Zhou Zhiyu

Amidst thousands of calls, Tencent, which has taken the lead in domestic social communication, has finally begun to extend its reach to independent AIGC apps. The battlefield is already full of waves, but no winners have come out yet.

On May 30, Tencent announced the official launch of the app “Tencent Yuanbao” based on the mixed-yuan model. Previously, Tencent also built mixed element assistants for the C side, but only the applet and PC side. Yuanbao is considered the first AIGC pure application built by Tencent for the C side.

Liu Yuhong, vice president of Tencent Cloud and head of Tencent's hybrid model, said that Tencent will definitely be ahead of time to expand its model.

Although the big model battlefield is very intense today, the penetration rate of AIGC products is very low, less than 1%, and the industry is still in its infancy.

Also, according to QuestMobile statistics, as of March 2024, the number of users in the AIGC app industry only surpassed 73.8 million. The top three apps Douyin Doubin, Baidu's Wenxin, and Tiangong, a subsidiary of Kunlun Wanwei, had monthly active users of 23.282 million, 14.661 million, and 9.661 million respectively.

Liu Yuhong pointed out that the first-mover advantage of ToC AIGC products is certainly important, but only if the product is mature and can successfully seize the position, but judging from the reality of a penetration rate of only 1%, this first-mover advantage is a bit insignificant.

In the same way that the mixed-yuan model was only brought to the front of the stage in September of last year, Tencent was also very careful in releasing the AIGC app. It decided to push the switch after refining the technology in Tencent's massive business scenarios and gaining insight into actual needs through a large number of users.

As a latecomer, Tencent hopes that Yuanbao will become an “all-around AI assistant” in everyone's pocket.

For work efficiency scenarios, Tencent Yuanbao provides core capabilities such as AI search, AI summary, and AI writing. Facing everyday scenes, Ingot is also very rich in gameplay, and provides a variety of special AI applications, such as versatile AI avatars, oral language sparring, and super translators. At the same time, gameplay such as creating personal agents has also been added.

Currently, large model applications are still in a very early stage, and there is a large “delay” from model capability to application implementation. According to data provided by Tencent, when people currently use products related to large models, more than 65% of the demand is focused on work/learning efficiency scenarios, but related AI product solutions are not yet mature.

In response to the three core requirements of the efficiency scenario: information acquisition, processing, and production, Tencent Yuanbao has focused on commercialization exploration.

In terms of AI search, Tencent Yuanbao is connected to search engines such as WeChat Search and Sogou Search, and is enhanced through AI search to improve the effectiveness of current and knowledge questions, making it more efficient than traditional searches. At the same time, the content covers Tencent ecological content such as WeChat accounts and authoritative Internet sources, and the accuracy of answers is higher.

Judging from the AI's ability to search, Yuanbao is also like a connector, linking all of Tencent's internal ecosystems, including WeChat public accounts, corporate WeChat, etc.

Liu Yuhong revealed that AI search is a relatively unique and well-done piece of Yuanbao, and further enhancing AI search capabilities is currently a key task for the mixed yuan team.

In terms of AI summarization, Yuanbao can upload up to 10 documents in various formats such as PDF, word, txt, etc., and can parse multiple WeChat public account links and websites at once. It supports 256K native window context, which is equivalent to a copy of “The Three Kingdoms” or the full English version of “Harry Potter”.

Based on this ability, whether you want to quickly understand a book or a new field, or deal with complex and complicated reports and documents, Ingot can help.

In terms of AI writing, Yuanbao not only supports multiple rounds of Q&A, but can also organize the content of conversations into reports and perform structured output according to requirements, greatly improving the efficiency of information from acquisition to processing to production.

From a functional point of view, there doesn't seem to be much difference between Tencent Yuanbao and existing AIGC applications on the market. What really determines the competitiveness of its products is the underlying large model capability.

Since its debut in September 2023, the parameter scale of Tencent's mixed-element model has been upgraded from 100 billion to trillion, the pre-training corpus has been upgraded from 1 trillion to 7 trillion tokens, and the first to a multi-expert model structure (MoE). The overall performance has improved by more than 50% compared to the Dense version.

According to reports, some of the Chinese proficiency of the mixed yuan model is on par with GPT-4, and their ability to answer “current and new” questions has also improved in mathematics, reasoning, etc.

In terms of multi-modal capabilities, Tencent's mixed-element large model has also been improved. For example, in the field of raw images, Tencent's hybrid Wenshengtu infrastructure has been fully upgraded to the same DiT architecture as Sora; in the field of live video, Tencent mixed-element supports various video generation capabilities such as Wensheng video, TuSen video, graphic video, etc., and the video generation length can reach 16 seconds.

In a context where the underlying technology of the AI model is still being iterated, and the C-side commercialization path is still unclear, Tencent chose a path that focuses more on practicality, first serving Tencent's own business, then opening up big model capabilities to the industry through Tencent Cloud, while the C-side has slowed down. At present, AI's drive for Tencent's own business has been reflected.

As far as Tencent is concerned, the reason why Tencent is putting C-end apps on the table now is also because the entire industry has high expectations for killer apps. After all, this can bring a real second growth curve to a company. Perhaps to seize a new huge traffic entry, Tencent cannot be absent.

QuestMobile pointed out that judging from market performance, users are extremely enthusiastic about AIGC, and demand for related independent apps continues to grow. By March of this year, the number of users in the AIGC app industry had increased more than 8 times over the same period last year.

The industry is in full swing, but at present, there are also truly phenomenal AIGC apps being created, and no apps have taken over users' minds. Tencent, which has a large number of users, still seems to have an opportunity to stage a story that later takes the lead.

Disclaimer: This content is for informational and educational purposes only and does not constitute a recommendation or endorsement of any specific investment or investment strategy. Read more
    Write a comment