It's called "Earthquake".
This morning, Jiahui Yu, a core researcher at Meta Superintelligence Labs, MSL, announced his resignation on X and plans to start a new company.
This is just eight days away from the release of Muse Spark 1.2.
"Building the TBD Lab together with Mark Zuckerberg and Alexander Wang has been an inspiring and rewarding experience", the family wrote on social media, saying they were proud of their multi-modal team's achievements on Muse Spark, Voice Mode, Muse Image and Muse Video.
The name, team and financing of the new company has not been disclosed, and as for the specific direction of the new company, he hung up the appetite of the outside world with one sentence:
"A question that will have a profound impact on the future of humanity, which very few people explore at the moment, is becoming more and more attractive to me.
According to the study paper platform alphaXiv, as of now, Meta has lost more than 200 prominent researchers.
Just after the new model was delivered, the core leaders turned to entrepreneurship
Looking back at the rest of his research career, "multi-modeling" is a constant theme.
Prior to the Big Model, his research focused on computer vision and image generation. After joining Google, he participated in visual generation projects such as Parti, and after the merger of Google Brain and DeepMind, became one of the core members of Gemini Multimodal Directions.
At the end of 2023, he transferred to OpenAI and became the head of the Perception team, where he was involved in the development of models such as GPT-4o, o3o, o4-mini, enabling models to process visual information further from "seeing images" to "thinking with images".
In June 2025, the family joined the newly formed Meta Superintelligence Labs, together with Mark Zuckerberg, Alexander Wang, to build the TBD Lab and be responsible for its multi-modal direction.
For the past year, he has been in charge of the multidimensional direction of TBD Lab. Just before leaving, the team had just completed a round of intensive model releases.
In August, Meta released Muse Spark 1.2. Compared to previous versions, the new model continues to enhance Agent's capabilities and multi-modal interaction, and is the core model behind new products such as Muse Code.
Meanwhile, Muse Image is now online, and Muse Video is in the works, as well as Voice Mode, the multi-modal version of TBD Lab, which is specifically mentioned in the rest of the company's resignation statement, is rapidly spreading.
But just over a week after the release of Muse Spark 1.2, the core executive decided to leave Meta.
The rest of the family was a key target of last year's "The Battle for Talent", but it was only 14 months after she joined the group.
It's time to move from the "Battle of the Seizures" to the "Stranger Stage".
More than a year ago, the outside world was still debating how many people Meta can dig from OpenAI, Google and Anthropic.
Now, the topic becomes how long these people can stay on Meta.
In May this year, Meta launched a large-scale personnel adjustment.
Redundancies, restructuring, and performance audits were tightened, and a large number of engineers were reassigned, many of whom switched from software development to data labeling.
This internal adjustment, called "efficiency", has reduced costs in the short term, but has accumulated a pervasive sense of insecurity within the organization.
Tech industry author Gergely Oros uses an iceberg map to describe Meta's current engineer status:
About 10 percent of those fired are looking for work; about 20 to 30 percent of those reassigned have decided to leave internally; the remaining 60 to 70 percent are not active but still applying for resumes.
They fear that they may be the next person to be tamed or judged at any time.
Gergely Orosz posted that starting in May, almost all Meta engineers were in a "shields down" state. They began actively responding to hunters and accepting external interview invitations. Anthropic, OpenAI, Google also took the opportunity to attract large numbers of people from Meta, targeting mainly senior engineers, field experts and core developers.
By August, with more and more engineers resigning, Meta began retiring talent with high bids.
Russian gives an example. An engineer who was assigned to an internal data labeling was mentally crushed and said, "I'm really pissed off". He went to a Google interview, and Google's initial offer was not high, but he also wanted to leave as soon as possible.
Upon learning that he was about to resign, Meta suddenly issued a counter-offer for a substantial raise, and the engineer took the offer back to Google to continue negotiations, with Google increasing his salary to slightly above Meta's retained price. Eventually, he left Meta with a higher salary.
According to Orosz, it is true that some people accepted the offer and chose to stay. However, seven of the recipients of the offer he communicated directly with eventually left. Similar situations have occurred in Menlo Park, New York and London.
Some netizens commented on Meta's current situation: in the "efficiency era" of continuous reduction and mobilization of personnel, into the "expatriate era", before the trust of the branch began to concentrate on settlement.
For Meta, model gaps may take months to catch up, and repair of organizational trust usually takes longer.
The link is as follows: