Economic Observer reporter Ren Xiaoning

On June 5th, Tang Daosheng, Senior Executive Vice President of Tencent Group and CEO of the Cloud and Smart Industry Business Group, had a nearly hour long conversation with Yao Shunyu, Chief AI Scientist of Tencent, which attracted attention from the outside world. This conversation took place at the Tencent Cloud AI Industry Application Conference. Several attendees stated that they mainly came for Yao Shunyu. The main forum on site was crowded with people, and there were no seats in the live broadcast room where the forum was broadcasted.

Yao Shunyu is a former OpenAI researcher who joined Tencent in December last year. After Yao Shunyu joined, Tencent's AI business underwent significant adjustments, including the establishment of AI Infra, AI Data, and Data Computing Platform departments, as well as adjustments to the AI Lab that had been established for nearly a decade. The latest big model Hy3 preview released by Tencent is the core achievement of the hybrid team led by Yao Shunyu.

After nearly an hour of conversation, Tang Daosheng was interviewed by media including the Economic Observer. Tencent's hybrid big model, as well as AI products such as Yuanbao, WorkBuddy, Qclaw, etc., all belong to the cloud and smart industry business group. Tang Daosheng is the overall person in charge of these businesses.

Previously, Tencent Hybrid was very concerned about external rankings, but after Yao Shunyu's arrival, it became based on user experience as the indicator. Previously, Tencent had a lot of low-quality data that was not helpful or even harmful for model training. After Yao Shunyu arrived, he cut off these harmful data and improved the quality of the data.

Tang Daosheng believes that Yao Shunyu has done a lot of things to simplify complexity, and the Hy3 preview model has made great progress compared to before. Yao Shunyu has made great contributions. He also believes that after Yao Shunyu arrived, the model and product Co Design brought great value to Yuanbao.

The main models integrated on Yuanbao are hybrid and DeepSeek. Currently, 80% of Yuanbao's users are using Hy3 preview, resulting in a significant increase in product retention rate. The cooperation between Yuanbao and Hunyuan is also becoming closer, and recently some departments of the two businesses have moved to the same building.

The biggest investment in Tencent's AI products is still Yuanbao. Tang Daosheng said that chatbots are a common demand and an important track. The KPI of the Yuanbao team is user growth and retention rate improvement. What he is most looking forward to is that Yuanbao can do a good job in user experience, search services, introduce more real-time data, and answer users' questions in their lives.

In addition, Tencent will also allocate budget to CodeBuddy and WorkBuddy.

During the interview, Tang Daosheng also responded to external concerns about whether Tencent AI is slow, whether Tencent AI products are racing, as well as current hot topics such as bean bun fees and AI self-developed chips.

In the past six months, Tencent's stock price has been in a downward trend, and the outside world believes that Tencent has slowed down in its AI business. Tang Daosheng explained that Tencent's business model is very diverse, with many products in various tracks, and many teams promoting different projects. In such a complex organization, some things may be done too fast, while others may be done too slowly.

After the crayfish craze this year, Tencent launched multiple agent products, with over 20 efficiency intelligent agent tools alone, which are considered to be conducting internal horse racing. Tang Daosheng replied that Tencent's AI products are definitely not racing horses. Although there are many scenarios for agent services, users' needs are also diverse. Tencent's AI products and intelligent agent products have their own positioning.

Regarding the charging of tofu buns, Tang Daosheng said that the commercialization of competitor products is ahead of Tencent. However, for Tencent, the most important thing at present is to provide a good product experience, find Yuanbao's differentiated positioning, and serve more users well.

At present, the Internet giants' self-developed AI chips are also the focus of attention. Alibaba and Baidu have self-developed chips. Tencent announced the progress of video chips in May this year, but did not announce AI chips. Tang Daosheng said that doing chip design alone cannot solve the production capacity problem, and Tencent's approach is to cooperate with more chip manufacturers.