Hao Zhang, an assistant professor at the University of California, San Diego (UCSD) and AI systems researcher, announced on Wednesday that he has officially left his position at Meta Superintelligence Labs' TBD Lab. Earlier this year, he had applied for academic leave to join Meta TBD Lab, and he is now ending this tenure to embark on a "new mission."

In a post on social media, Hao Zhang described this as "the best summer I have had in years," and specifically mentioned one of the team leaders, @yinghai, as "one of the best infrastructure leads, with deep technical skills and ambitious goals, and very adept at creating an environment for the team to focus on R&D." He also highlighted his experience working with @_tim_brooks (Tim Brooks, whom he referred to as "one of the greatest minds who pioneered the entire video generation industry"). Hao Zhang stated that he contributed to the development of Meta's Muse image and video generation model's inference capabilities and witnessed its rapid scaling, praising the recently released consumer AI product Muse as "one of the most impressive consumer-grade intelligent agent products currently available."

Hao Zhang's departure is not an isolated case but the latest in a series of high-profile exits from Meta Superintelligence Labs since its establishment this year. According to previous reports, within about two months of Zuckerberg announcing the formation of the lab, at least eight employees had left, including Chaya Nayak, who had worked at Meta for ten years (now at OpenAI), and Birch Mart, who had been with the company for twelve years (now at Anthropic), with at least two others returning to OpenAI.

Another notable departure occurred around September 9: Andrew Tulloch, a core member of TBD Lab, which is considered central to Zuckerberg's AI strategy and led by Scale AI founder Alexandr Wang, announced his departure. Tulloch was a star researcher personally recruited by Zuckerberg from Thinking Machines Lab, and reports indicated that his compensation package was as high as $1.5 billion over six years (this figure is disputed, but even if discounted, it remains a rare high-salary contract in the tech industry). Tulloch specifically waited until after Meta successfully launched its new generation of open-source AI models and the consumer AI assistant Muse before officially leaving; his next steps and whether he will continue AI research at another institution are currently unclear.

Reports indicate that despite Zuckerberg offering "nine-figure" compensation packages to attract top talent, this aggressive pay strategy has also caused dissatisfaction among long-time Meta employees, being seen as one of the factors leading to a tense team atmosphere and ongoing talent loss. The reports quote opinions suggesting that the company needs to make substantial improvements in the work environment to truly retain talent. Meta has responded to this wave of departures with indifference, stating, "Some people choose to stay in their original positions rather than jump to new companies... this is normal."

From the talent market perspective, top AI research talent is currently in extreme demand, with competition among leading institutions like Anthropic, OpenAI, Google, SpaceX AI, and Meta entering a model of "sky-high salaries + short-term high-intensity sprint projects"—many researchers choose to participate in key projects at popular labs in a manner similar to Hao Zhang's "academic leave + short-term joining," and after achieving phase results (such as the release and rapid scaling of Muse), they opt to leave, returning to academia or other institutions. This flow model benefits individual researchers (allowing them to accumulate experience from top projects and receive high compensation) and employers willing to exchange high salaries for short-term explosive output; however, for employers, this also means that the retention period of core team talent is significantly compressed, posing a continuous risk of losing organizational knowledge and technical continuity.

It is noteworthy that Hao Zhang's departure occurred shortly after Meta successfully launched the new generation of open-source AI models and the consumer AI assistant Muse, and it bears a striking resemblance to Andrew Tulloch's timing of leaving "after the product launch," a pattern that has repeatedly appeared in recent departure cases from TBD Lab.

Source: Public Information