At the beginning of this year, Tencent's AI assistant application "Yuanbao" underwent a significant organizational restructuring, transitioning its product team from the Technical Engineering Group (TEG) to the Cloud and Smart Industry Group (CSIG). Following this change, the responsibility for Yuanbao's product capabilities and user experience optimization has been assigned to Wu Zulong, who is also the head of Tencent Meetings.

According to Tencent's official classifications, the "Hunyuan" series consists of the Hunyuan large model along with Yuanbao and YuanqiYuanbao is positioned similarly to Doubao, while Yuanqi is compared to KouziEvaluating their performance, Tencent's AI applications had a moderate showing last year, deviating from the ideal of "user-centric" and "product experience first," which the company espousesThis raises questions about the company's genuine commitment to these principles.

The hesitance surrounding Tencent's large models and AI capabilities can be contrasted starkly with ByteDance's decisive approach to technological advancementTencent's actions have often come off as reactive, with moves like the Hunyuan model and Yuanbao feeling more like forced reactions instead of proactive initiatives.

From a business perspective, some insiders assert that Yuanbao essentially serves as a technology demonstration model that is not fully developed as a user-facing productThis separation of Yuanbao from the technical department marks a pivotal shift towards valuing the product's experience and usability.

It is anticipated that Yuanqi will also be integrated into CSIGWith a focus on B2C, it has the potential to evolve into an independent agent application, while for B2B it can be utilized in enterprise scenarios through the combination of "large models and agents."

Notably, Wu Zulong has extensive experience in B2B markets while also being able to cater to consumer-facing applicationsThis might signal a significant repositioning for Yuanbao that mirrors the successful path taken by Tencent Meetings, aiming to enhance consumer experience while directly addressing enterprise client needs in the context of AI commercialization.

In the ongoing competition in AI and large model development, the backing of Tencent's WeChat ecosystem and social traffic is a formidable force that cannot be overlooked

Advertisements

The company has previously entered late into emerging trends, capturing a critical opportunity during the rise of short videos with its "Video Number." This past success leads to the question: can history repeat itself?

Tencent has frequently found itself lagging in the era of large modelsFor instance, the release of the Hunyuan large model in September 2023 came five months after Alibaba's Tongyi Qianwen and nearly six months behind Baidu's Wenxin Yiyan.

Like Meta's strategy abroad, Tencent is attempting to bridge this gap through open-source initiatives to facilitate rapid model iterationBy last year, it had already released several open-source models, including text generation, graphics generation, 3D generation, and video generation schemes.

Tencent's cautious stance towards new technologies often reinforces a follow-lag approach, where the company waits for a clear community directive before diving headfirst into developmentsThis cautious investment behavior, as seen with its late entry into investing in companies like "The Dark Side of the Month" only after they reached "first-tier" status, reflects a hesitance that contrasts sharply with the aggressive expansion strategies adopted by rival companies.

The leadership's prioritization of measured approaches is quite telling; while ByteDance’s Zhang Yiming is known for burning the midnight oil on research papers to engage directly with top talent in academia, Tencent has continued to stress the importance of grounding technology initiatives in practical business applications.

During the 2024 staff meeting, Ma Huateng reiterated the importance of focusing on core business, reducing costs, and rejuvenating existing sectorsCritics argue that Tencent's approach, reminiscent of its internet-era mindset, is about minimizing resource commitment while passively observing competitive shifts before choosing to engage when conditions appear favorable for a comprehensive assault on the market.

This strategy has significant implications for the product front, where Yuanbao's absence from the previous year’s investment flow battles underscored its underwhelming reach

Advertisements

As of December last year, Yuanbao's app reported only 2.91 million monthly active users, a mere fraction of Doubao’s engagement figures.

Internally, various smaller teams within Tencent are engaging in a "horse race" to spur innovative applications based on the Hunyuan large model architectureFor example, the recently launched smart workspace product, "ima.copilot," is one outcome of this energetic innovation push.

In the origin story of the Hunyuan Assistant, it had the potential to be a stellar project, spearheaded by Zhang Zhengyou, who holds the highest professional rank in Tencent's historyThe initiative attracted key figures from TEG, CSIG, PCG, and personnel from both the WeChat and interactive entertainment sectors.

Despite the star-studded cast, the "Hunyuan" series has struggled to gain traction within the industryAmong the foremost challenges is the talent deficit problem; unlike ByteDance's aggressive acquisition of talent and research expansion, Tencent AI Lab, tied to TEG, suffered significant layoffs in 2022 that contributed to a perception of insufficiency among employees.

Moreover, reports indicate that many members of the Hunyuan team originated from search promotion backgrounds, leading to worries about their capability to innovate effectively in this spaceThe lack of a clear direction, with some employees describing the environment as chaotic and lacking definition, has further contributed to dissatisfaction.

With ongoing personnel losses and reports of key tech leads, such as Liu Wei, departing to pursue ventures in video generation, there is growing concern regarding the sustainability of Yuanbao and Yuanqi's developmentUsers lament the scarcity of high-quality data, especially from WeChat's wealth of search resources, within the crowded AI assistant application landscape.

Liu Yuhong, the lead for the Hunyuan large model, has previously emphasized that enhancing user experience is paramount—focusing initially on fulfilling functional needs characterized by enhanced search features while simultaneously aiming to integrate more deeply into the WeChat ecosystem, enhancing the application’s distinctiveness.

As of now, Yuanbao has begun integrating with other ecosystem products, including Tencent Docs, PC Manager, and search input methods; however, the Hunyuan series is in great need of a transformational figure akin to "Zhang Xiaolong" to exert substantial product influence

Advertisements

In Tencent’s history, the company has been a fertile ground for product managers, influencing numerous AI applications, such as Kimi, which continue to thrive on Tencent's product logic.

The executives may have identified Wu Zulong, who pioneered the Tencent Meetings project from concept to full execution, as a suitable leaderTencent Meetings stands out as a phenomenon with over 400 million users as of September 2023, achieving a five-fold increase in paid meeting numbers when compared year-on-yearWhile it boasts a significant consumer base, the initial conception remains deeply rooted in the B2B sphere.

In a gesture similar to the restructuring of Alibaba's Tongyi within its larger organizational infrastructure, redirecting the leadership towards a more consumer-focused approach, Tencent's strategy seems to aim for a seamless transformation towards B2B commercialization.

If this is the case, then Yuanbao's integration into CSIG undoubtedly reflects Tencent's investment in setting up a viable path for future monetizationThe primary similarity between Yuanbao and Tencent Meetings revolves around the synergy between C-end consumer experiences and B-end commercialization pathways.

Wu Zulong has previously indicated that addressing consumer challenges in various scenarios leads to the gradual sedimentation of new capabilities and use cases, which can subsequently be cycled back into enhancing the product itself, ultimately becoming a significant service for organizations and enterprises.

After a year-long period of observance, Tencent may have finally recognized the commercial viability of AI applications, prompting them to seek to replicate the success achieved by Tencent Meetings and create strong bridges between Yuanbao's functions for both B2B and B2C marketsTencent Meetings continues to lead in merging scenarios with innovative technologies, and underpinned by the Hunyuan large model, has ushered in features such as intelligent recording, automatic summarization, and the Tencent Meetings AI assistant.

As we advance into the new year, Ma Huateng has outlined an expectation for TEG to fully embrace the productization of large models across its various business groups

Products like WeChat, QQ, input methods, and browsers are gearing up to unveil AI capabilities, with gaming, WeChat Reading, and Tencent Video among those exploring more opportunities with the Hunyuan framework.

In the competitive industry landscape, Tencent's potential traffic leverage remains underappreciatedIn stark contrast, platforms like Douyin and Bilibili, which aggressively built momentum for applications like Doubao and Kimi, still see Yuanbao struggling to harness the combined power of "WeChat+" effectively, which could dramatically alter the competitive dynamics of the space.

Tencent has previously executed an impressive turnaround with its Video Number initiativeLaunched officially in 2020, it has dramatically evolved through consistent iterations, unleashing over forty new features and elevating daily active user counts, to the point where it is now prominently featured in Tencent’s financial reports as a cornerstone for the company’s growth.

Tencent's cautious yet arguably prudent approach in the large model domain, potentially refined by previous successes, allows them to maneuver without undue hasteHowever, sustained technical commitment to large models necessitates an accumulation of resources over timeReports suggest that in 2024, ByteDance may emerge as NVIDIA's second-largest client, ordering approximately 230,000 chips, with Tencent following close behind.

The market observes a wave of consolidationSome companies have chosen to scale back on large pre-trained models focusing on narrower goalsThis restraint exhibited over the previous phase may have armed Tencent with a significant store of ammunition, granting them an edge to transition from defense to offense.

On one end, it is evident that the competition within the large model sphere has narrowed primarily to major players, with the capacity to continually inject resources and computational strength typically reserved for powerhouses like Tencent, Alibaba, ByteDance, and Baidu

Advertisements

Advertisements