Zhang Yiming's Remarks Inspire ByteDance CEO's Strategy: Acknowledging Short-Term LLM Gaps, Committing to Long-Term Self-Development

Deep News
08/06

Following founder Zhang Yiming's recent comments, ByteDance CEO Liang Rubo has outlined the company's future goals for its large model business. On August 6, it was reported that ByteDance held its mid-year all-hands meeting for 2026, where CEO Liang Rubo and other executives addressed the company's direction.

At a recent internal meeting, ByteDance founder Zhang Yiming stated that developing models requires a commitment to long-termism and delayed gratification, rather than using others' outputs in exchange for temporary rankings. During the all-hands meeting, Liang Rubo summarized the performance of ByteDance's AI business in the first half of 2026: the Doubao app remained competitive in the consumer market, and the video generation model Seedance achieved SOTA (State Of The Art) status. However, the company acknowledged that its large language model (LLM) has fallen behind leading overseas models.

Liang Rubo stated that for large language models, ByteDance will continue to insist on self-development, focusing on getting the fundamentals right. He emphasized accepting short-term setbacks while consistently optimizing for the long term.

Liang Rubo also explained the background behind the recent integration of Doubao, Feishu (Lark), and Volcano Engine. Previously, ByteDance announced a restructuring of its AI-related businesses, creating a new "Creativity Service Platform" department at the enterprise service level. This aims to integrate ByteDance's B2B capabilities to better meet enterprise clients' AI needs. Liang Rubo noted that AI's development in productivity has exceeded expectations, making the B2B sector even more critical. He revealed that over 90% of new Feishu clients are simultaneously purchasing Feishu's AI products. This integration is a strategic preparation to create superior products for office and productivity scenarios and better serve B2B clients in the face of AI's vast opportunities.

When an employee asked how to interpret his earlier mention of "narrowing business scope," Liang Rubo responded that the company has indeed been concentrating resources on "high-priority" businesses over the past few years, divesting some non-core operations. He explained that the company's business strategy formulation has recently been summarized as "high priority, sturdy trunk, optimize for the long term." "High priority" refers to creating incremental social value—doing things better than others can, making the whole society more efficient. "Sturdy trunk" means focusing energy on a few exceptionally important and closely related matters. Liang Rubo introduced that ByteDance's core businesses include three areas: AI, Information Platforms, and Transaction Services. These are closely interconnected, all using computation to enhance intelligence to fulfill the company's mission of "Inspire Creativity, Enrich Life." Furthermore, "sturdy trunk" implies that the business itself is large and important, and can also drive and support other businesses. For example, Douyin (TikTok) as a "trunk" business helps e-commerce and local services. He expressed hope that Doubao will also become a "trunk" business in the future.

Regarding the 2026 annual keyword "Climb the Peak," proposed to enhance AI technical capabilities and integrate existing products and services through Doubao/Dola Assistant, Liang Rubo stated that "optimize for the long term" means being user-centric and focusing on incremental social value, without being swayed by short-term fluctuations. For the AI business, this means believing in AGI and insisting on self-development.

Tan Dai, the newly appointed head of the "Creativity Service Platform" department, cited the example of Seedance 2.0. He noted that in the B2B field, only SOTA models can first unlock scenarios with high economic value. Before this model, video generation was "more of a toy," but Seedance 2.0 has pushed the technology towards real application in serious and commercial content production. He revealed that Seedance's usage volume is much higher on workdays than weekends, indicating its primary application is in work production and creation.

Addressing the gap between ByteDance's LLM and overseas SOTA models, Tan Dai said the most important task now is to get the basics right, aiming to enter the first tier over the long term without taking shortcuts. Liang Rubo added that it is crucial to maintain correct execution and not deviate from the strategy. "We must psychologically accept being behind for a period of time in large language models, insisting on self-development and optimizing for the long term. I saw an external report yesterday speculating that we are insisting on self-development due to external pressure. That's a guess. The real reason is that we want the team to practice delayed gratification and build a solid technical foundation, which is key to achieving AGI in the long run."

Earlier, on June 29, Liang Rubo sent an all-staff email updating ByteDance's culture and management philosophy, including leadership principles. External attention was drawn to updates like "dare to set high goals" and "go deep into the front line," with some interpreting the inclusion of leadership principles in performance reviews as "cutting middle management." In response at the all-hands meeting, Liang Rubo clarified that these principles are not entirely new, but a summary of thoughts from the past few years. The update was prompted by the new wave of AI opportunities and challenges faced by the organization at its current stage, aiming to clarify "how we should proceed."

When asked about "standards for top talent in the AI era," Liang Rubo stated that the most important quality is the ability to create value using AI. Top talent needs to be able to discover value, define problems, coordinate resources, and create value, fully leveraging AI to enhance efficiency throughout the process. From the perspective of AI's development, the last role to be replaced by AI is the "Organizer," so individuals should enhance their skills as an Organizer.

To help employees adapt to changes, ByteDance offers not only AI tools and token support for office scenarios but also a reimbursement mechanism for personal AI learning experiences. The company dynamically adjusts the reimbursement cap through periodic assessments to support employees' skill upgrades. Currently, ByteDance is conducting its campus recruitment for the 2027 graduate cohort. Regarding talent development for new graduates, the relevant person in charge stated that campus recruits are becoming an increasingly important and growing source of talent. The company will continue to increase investment in campus recruitment. For the development of campus hires, the company will strengthen guidance through necessary training and mentorship, but most importantly, place them in suitable roles to learn by doing and fully realize their potential, offering young people more opportunities to take on key tasks. According to data disclosed at the all-hands meeting, campus hires outperform the company average on core metrics such as performance, promotion, and retention. For example, since 2025, the proportion of campus hires receiving M+ or higher ratings in their second performance review has been significantly higher than the company average. Their three-year turnover rate is significantly lower than the company average. At the same job level, campus hires experience faster promotions and higher promotion frequency.

免责声明:投资有风险,本文并非投资建议,以上内容不应被视为任何金融产品的购买或出售要约、建议或邀请,作者或其他用户的任何相关讨论、评论或帖子也不应被视为此类内容。本文仅供一般参考,不考虑您的个人投资目标、财务状况或需求。TTM对信息的准确性和完整性不承担任何责任或保证,投资者应自行研究并在投资前寻求专业建议。

热议股票

  1. 1
     
     
     
     
  2. 2
     
     
     
     
  3. 3
     
     
     
     
  4. 4
     
     
     
     
  5. 5
     
     
     
     
  6. 6
     
     
     
     
  7. 7
     
     
     
     
  8. 8
     
     
     
     
  9. 9
     
     
     
     
  10. 10