On June 16, 2025, Li Hang, the director of ByteDance's AI Lab, officially transitioned from his leadership role to a consultancy position within the company.

A ByteDance spokesperson clarified that reports of "Li Hang's departure" were misleading; he has, in fact, retired and been rehired as a consultant, maintaining his previous scope of work. Born around 1966, Li Hang may have reached the standard retirement age of 60.

The AI Lab serves as ByteDance's hub for AI research. Established in 2016 under the leadership of former Microsoft executive Ma Weiying, who reported directly to Zhang Yiming, it peaked at 150 employees and spearheaded innovations in recommendation algorithms and large models.

However, ByteDance's AI strategy has evolved in recent years. In 2020, the AI Lab shifted from a forward-looking group-level project to a technology middle platform supporting various internal businesses.

In February 2025, Wu Yonghui, a former vice president of Google DeepMind, joined ByteDance to head the large model Seed team. Li Hang, who succeeded Ma Weiying as the AI Lab's first leader, subsequently reported to Wu Yonghui.

Beginning in 2023, the NLP team responsible for large language models and PixleDance, which developed video generation models, were transferred to the Seed team. By April, the restructuring was complete, with media reports indicating that the ByteDance AI Lab had fully integrated into the Seed team.

As the race to develop large models intensifies, companies are hurrying to integrate technology into products. ByteDance is competing with OpenAI, Google, and others to expedite product launches and capture market share. While the AI Lab previously focused on academic research, it now requires individuals with stronger business acumen to drive commercialization. Industry analysts suggest that Li Hang, as an academic expert, may not be as adept as business-oriented talent in technology implementation.

Notably, since Wu Yonghui took charge of ByteDance's large model team in February 2025, three out of the four key figures under the previous leader Zhu Wenjia – Li Hang, Yang Jianchao, Qiao Mu, and Xiang Liang – have undergone changes.

First, Qiao Mu, head of the large language model team, was suspected of having his Feishu account deactivated due to an extramarital affair.

Subsequently, rumors circulated that Yang Jianchao, head of vision multimodal generation, was planning to leave the company (unconfirmed). It is speculated that Yang Jianchao may have been considering resignation due to work pressure, time differences with North America, and other factors. Zhou Chang, a former technical leader at Alibaba Tongyi, is rumored to be a potential successor.

Now, Li Hang has also stepped down as AI Lab director, retired, and returned as a consultant.

About Li Hang

Li Hang graduated from the Department of Electrical Engineering at Kyoto University in 1988 and earned a Ph.D. in Computer Science from the University of Tokyo in 1998.

His professional journey is extensive. From 1990 to 2001, he served as a researcher at NEC Corporation's Central Research Laboratory in Japan. Between 2001 and 2012, he worked at Microsoft Research Asia, progressing from researcher to senior researcher and principal researcher. In 2012, he joined Huawei Noah's Ark Lab as chief scientist and lab director.

In 2017, Li Hang joined ByteDance's Artificial Intelligence Laboratory, leading the team in foundational and applied machine learning research. In 2022, he published a paper on the Transformer architecture, which became a cornerstone of GPT series models. The AI Lab under his leadership pioneered advanced research in artificial intelligence, encompassing natural language processing, robotics, AI for Science, and other domains.