Mark Zuckerberg announces Meta Superintelligence Labs — with a battalion of AI gurus poached from OpenAI, Google, and DeepMind to try and secure an AGI win

by

For the last several weeks, it appears that Meta has been aggressively pushing forward, striving to keep pace and potentially surpass competitors such as Google, OpenAI, Anthropic, and Microsoft, who seem to be significantly ahead in the AI field.

Initially, Mark Zuckerberg committed $14.3 billion towards Scale AI and appointed Alexandr Wang as a joint leader for the operations in Meta’s advanced artificial intelligence research facility, which he will share with Nat Friedman, who was previously the head of GitHub.

News started circulating, suggesting that Mark Zuckerberg was aggressively recruiting top AI specialists from competing companies, aiming to give Meta an edge in the competitive AI arena. It is said that this high-level recruitment came with a hefty signing bonus of up to $100 million.

Instead of using strict non-compete agreements backed by extensive paid time off to discourage employees from moving to competing firms, as Google’s DeepMind does, Mark Zuckerberg has successfully persuaded some top AI professionals to establish Meta’s new AI lab. This could potentially help Meta compete more effectively with its rivals in the field.

On Monday, Zuckerberg unveiled a company-wide communication to Meta employees, introducing the Meta Artificial Intelligence Research Center (as reported by Bloomberg). As stated by the executive:

With the rapid advancement of artificial intelligence, achieving superintelligence seems increasingly possible. I firmly stand behind my belief that this milestone marks the dawn of a fresh epoch for mankind, and I am unequivocally dedicated to ensuring that Meta remains at the forefront of this exciting journey.

Mark Zuckerberg has shared his intention to utilize the latest research facility to develop advanced AI systems that surpass human abilities. These AI models are intended to become an integral part of how individuals engage with technology.

Here’s the full internal memo as shared by Mark Zuckerberg to Meta staffers:

With the rapid advancement of artificial intelligence, achieving superintelligence is becoming increasingly possible. I firmly believe this marks the start of a new chapter in human history, and I am resolute in our quest to make Meta a pioneer in this field. Today, I’d like to shed some light on the structure of our AI initiatives, as we strive towards our goal: bringing personal superintelligence within everyone’s reach.

Let’s name our entire operation Meta Superintelligence Labs (MSL). This encompasses all our foundations, products, FAIR units, and a fresh laboratory dedicated to refining the upcoming generations of our models.

As a fellow researcher, I’m thrilled to share that Alexander Wang has joined our team at Meta, taking on the roles of Chief AI Officer and head of MSL. Over the years, I’ve had the privilege of collaborating with Alex, and I can confidently say that he is one of the most outstanding founders of his generation. His vision for superintelligence is profound, and under his leadership as co-founder and CEO, ScaleAI has rapidly grown into a dynamic force in our industry, contributing significantly to the development of nearly every leading model.

Nat Friedman has teamed up with Alex at Meta to oversee MSL, focusing on AI products and research. Nat will collaborate with Connor to clarify his future role within the company. In his previous stint, he managed GitHub under Microsoft, and more recently, led one of the top artificial intelligence investment firms. Nat has been a part of our Meta Advisory Group for the past year, giving him an understanding of our strategic direction and the tasks at hand.

In addition, we’ve welcomed some impressive new additions to our team recently, whom I’m thrilled to introduce.

Trapit Bansal — pioneered RL on chain of thought and co-creator of o-series models at OpenAI.

Shuchao Bi is a key contributor, having helped create the GPT-4 voice mode and o4-mini. In his previous role, he oversaw post-training efforts in multi-modal research at OpenAI.

Huiwen Chang is a key figure in the development of GPT-4’s image generation. Previously, she designed the MaskGIT and Muse text-to-image frameworks at Google Research.

I had the privilege of assisting in the construction of various models such as o3/o4-mini, GPT-4o, GPT-4.1, GPT-4.5, 4o-imagegen, and the Operator Reasoning Stack.

Joel Pobar has experience working with inferences at Anthropic. Before that, he was part of Meta for eleven years, where he focused on HHVM, Hack, Flow, Redex, performance tooling, and machine learning.

As a researcher spearheading the pre-training technology at Gemini, I am proud to have played a pivotal role in the development of our innovative projects, such as Gopher and Chinchilla. My work in these early language model initiatives at DeepMind has laid a solid foundation for Gemini 2.5’s advancements.

Hongyu Ren is the co-developer of GPT-4, 40-mini, 1-mini, 3-mini (various versions), and he has also worked on 3 and 4-mini models. Previously, he led a team specializing in post-training at OpenAI.

Johan Schalkwyk — former Google Fellow, early contributor to Sesame, and technical lead for Maya.

As a researcher in the field, I have recently been involved with post-training, coding, and reasoning tasks related to the Gemini project at Google DeepMind. Notably, I was instrumental in developing the latest iterations of Waymo’s perception models for the past two generations.

Jiahui Yu is a key contributor behind the development of models like o3, o4-mini, GPT-4.1, and GPT-4o. Prior to this role, she headed the perception team at OpenAI, and collaboratively led the multimodal team at Gemini.

Shengjia Zhao played a significant role in the development of ChatGPT, GPT-4, as well as various smaller versions (including 4.1 and o3). He previously oversaw the creation of synthetic data at OpenAI.

In this rephrased version, I aimed to maintain the original meaning while using simpler language for a more natural reading experience.

I’m thrilled about the advancements we have in store for Llama versions 4.1 and 4.2. These cutting-edge models drive Meta AI, a crucial component used by over a billion active users monthly across our apps, as well as a growing number of agents within Meta that enhance our products and technology. We’re dedicated to further developing these models.

At the same time, we’re embarking on research for our upcoming generation of models, aiming to reach cutting-edge advancements within the next year approximately. Over the last few months, I’ve been networking with leading professionals in Meta, various AI research institutions, and innovative startups to assemble a skilled core team for this compact yet potent initiative. We’re currently building this team and will invite numerous individuals from our broader AI organization to become part of this lab as well.

Meta stands out as an ideal candidate for introducing global superintelligence, given our robust business model capable of expanding computational resources beyond smaller entities. With a rich history of developing and scaling products that resonate with billions, we are at the forefront of the rapidly expanding AI glasses and wearables sector. Furthermore, our corporate setup enables us to act decisively and boldly in pursuit of our goals. I am confident that this fresh infusion of expertise and innovative methodology for model development will position us to fulfill the potential of personalized superintelligence for all.

More fantastic individuals are set to join our endeavor in the upcoming weeks, so keep an eye out! I’m thrilled to jump right in and start working alongside them.

Read More

2025-07-01 13:09