Jiahui Yu, a core researcher at Meta Superintelligence Labs, said on X on August 14 that he is leaving Meta and plans to launch a new company, ending a stint that began in late June 2025. His exit came eight days after his team released Muse Spark 1.2, the latest model in Meta's Muse family. In his departure message, Yu said, "Building TBD Lab alongside Mark Zuckerberg and Alexandr Wang has been an incredibly inspiring and fulfilling experience," and said he was proud of the multimodal team's work on Muse Spark, Voice Mode, Muse Image and Muse Video. He did not provide the new company's name, team details or funding status, saying only that he was being drawn to "a problem that will profoundly shape humanity's future, yet remains largely unexplored." Yu's move follows a career spanning several major AI groups. He studied at the University of Science and Technology of China's School of the Gifted Young and completed a PhD at the University of Illinois Urbana-Champaign under Thomas Huang, where he co-developed the image inpainting model DeepFill. He later worked at Google Brain, became part of the Gemini multimodal team after the Google Brain-DeepMind merger, and moved to OpenAI in late 2023 as head of the Perception team, contributing to GPT-4o, GPT-4.1 and the o-series models. He joined Meta's newly formed superintelligence team in June 2025 alongside former OpenAI researchers including Shengjia Zhao, Shuchao Bi and Hongyu Ren. His departure also revives attention on the aggressive pay packages Meta used to recruit AI researchers. *WIRED* reported at the time that select top-tier talent could receive packages worth up to $300 million over four years, with first-year total compensation potentially exceeding $100 million. Meta CTO Andrew Bosworth later said the widely circulated $100 million figure was not universal and reflected total package values including stock, bonuses and performance conditions, though Yu was still labeled the "$100 Million Man" by outside observers. At Meta, Yu was a founding member and multimodal lead of TBD Lab, the core model research team within MSL. The team released Muse Spark in April as MSL's first foundation model, signaling a shift from the Llama series to the Muse series for frontier models, then followed with Voice Mode, Muse Image and Muse Video. Muse Spark 1.2, released in early August, strengthened agent capabilities and multimodal interaction and became the core model behind products including Muse Code. Yu's exit comes against a broader backdrop of churn at Meta. alphaXiv said more than 200 prominent researchers had left the company before Yu, while its search page showed 929 researchers matching the criteria of having previously worked at Meta and no longer being there. Tech industry writer Gergely Orosz said that since May, nearly all Meta engineers had been in "shields down" mode, actively taking recruiter outreach and interviews. He estimated about 10% were laid off and openly job-hunting, 20% to 30% were reassigned and had internally decided to leave, and 60% to 70% were unaffected but still sending out resumes. The pressure followed Meta's May restructuring of layoffs, reorganizations and tighter performance reviews, which reassigned many engineers, including some from software development to data labeling. Orosz said Meta later responded with sizable counter-offers, but all seven counter-offer recipients he spoke with still left. Yu's departure is likely to intensify questions over whether Meta's recruiting strategy can produce durable retention as competition with OpenAI, Google and Anthropic continues.