Connect with us

AI Research

Key takeaways from Mark Zuckerberg’s memo on Meta’s ‘Superintelligence’ AI push and big hires

Published

5 days ago

on


Meta CEO Mark Zuckerberg announced a major overhaul of Meta’s AI efforts in a memo to staff on Monday, including a clutch of significant hires from rival AI companies.

The restructuring comes after Meta has struggled in recent months to stay at the cutting edge of AI. Zuckerberg has largely staked Meta’s future on the rapidly-evolving technology, saying the company will spend between $64 billion and $72 billion building out data centers to handle AI workloads this year. This is up from just $28 billion in annual capital spending in 2023.

Zuckerberg’s announcement signals a major strategic shift and aggressive investment in AI, with the CEO reportedly investing billions of dollars to secure key AI talent. Here are key takeaways from Zuckerberg’s memo and background on the rationale behind Meta’s moves:

  • Creation of Meta Superintelligence Labs (MSL):
    Meta is consolidating all its efforts that involve building large AI models—including its teams working on its Llama models, product teams, its Fundamental AI Research (FAIR) team, as well as a brand new unit focused on developing the next generation of cutting-edge AI—under a new division, MSL. MSL is being co-led by ex-Scale CEO and cofounder Alexandr Wang, who is becoming Meta’s “Chief AI Officer” and ex-GitHub CEO and AI investor Nat Friedman.
  • Strategic Goal:
    The explicit aim is to build “personal superintelligence for everyone,” Zuckerberg said in the memo. Superintelligence would be AI that can perform beyond human level at most cognitive tasks.
  • Aggressive Talent Acquisition:
    Meta is poaching top AI talent from rivals like OpenAI, Google, and Anthropic. To do so, he is offering unprecedented signing bonuses—up to $100 million, according to comments made by OpenAI CEO Sam Altman. In his memo, Zuckerberg announced the hiring of 11 well-respected AI researchers from these other AI labs. Meta also invested $14.3 billion into Scale AI as part of the deal to bring Wang to Meta and reportedly invested billions into Friedman’s AI-focused venture capital firm to secure his move to Meta.
     
  • Failed Acquisition Attempts:
    The hiring spree follows rebuffed attempts to acquire key AI startups, including former OpenAI Chief Scientist Ilya Sutskever’s company Safe Superintelligence and Perplexity AI.
  • Llama Struggles and Competitive Pressure:
    Meta’s latest Llama AI model family, called Llama 4, has underperformed expectations. The company faced criticism that it published misleading benchmark figures for Llama 4, designed to make the models appear more competitive than they actually are. The release of Llama 4 Behemoth, the largest—and, according to Meta, the most powerful—model it has produced, has repeatedly been delayed. Meta has not yet debuted models with “reasoning” capabilities, losing ground to rivals such as OpenAI, Anthropic, Google, DeepSeek, and Alibaba’s Qwen. This has led to internal debate about Meta’s AI direction and increased urgency to revitalize its AI portfolio.
  • Retention and Reputation Challenges:
    Meta has suffered from the loss of key Llama researchers to competitors, further increasing the need to attract and retain world-class AI talent. The company also faces ongoing legal and ethical scrutiny over data practices,
  • Massive Capital Commitment:
    Meta is investing tens of billions in infrastructure, data centers, and custom hardware in an attempt to secure a leading role in the AI era.

What follows is the full-text as Zuckerberg’s memo, which was obtained by Fortune reporter Sharon Goldman:

As the pace of AI progress accelerates, developing superintelligence is coming into sight. I believe this will be the beginning of a new era for humanity, and I am fully committed to doing what it takes for Meta to lead the way. Today I want to share some details about how we’re organizing our AI efforts to build towards our vision: personal superintelligence for everyone.

We’re going to call our overall organization Meta Superintelligence Labs (MSL). This includes all of our foundations, product, and FAIR teams, as well as a new lab focused on developing the next generation of our models.

Alexandr Wang has joined Meta to serve as our Chief AI Officer and lead MSL. Alex and I have worked together for several years, and I consider him to be the most impressive founder of his generation. He has a clear sense of the historic importance of superintelligence, and as co-founder and CEO he built ScaleAI into a fast-growing company involved in the development of almost all leading models across the industry.

Nat Friedman has also joined Meta to partner with Alex to lead MSL, heading our work on AI products and applied research. Nat will work with Connor to define his role going forward. He ran GitHub at Microsoft, and most recently has run one of the leading AI investment firms. Nat has served on our Meta Advisory Group for the last year, so he already has a good sense of our roadmap and what we need to do.

We also have several strong new team members joining today or who have joined in the past few weeks that I’m excited to share as well:

  • Trapit Bansal — pioneered RL on chain of thought and co-creator of o-series models at OpenAI.
  • Shuchao Bi — co-creator of GPT-4o voice mode and o4-mini. Previously led multimodal post-training at OpenAI.
  • Huiwen Chang — co-creator of GPT-4o’s image generation, and previously invented MaskGIT and Muse text-to-image architectures at Google Research
  • Ji Lin — helped build o3/o4-mini, GPT-4o, GPT-4.1, GPT-4.5, 4o-imagegen, and Operator reasoning stack.
  • Joel Pobar — inference at Anthropic. Previously at Meta for 11 years on HHVM, Hack, Flow, Redex, performance tooling, and machine learning.
  • Jack Rae — pre-training tech lead for Gemini and reasoning for Gemini 2.5. Led Gopher and Chinchilla early LLM efforts at DeepMind.
  • Hongyu Ren — co-creator of GPT-4o, 4o-mini, o1-mini, o3-mini, o3 and o4-mini. Previously leading a group for post-training at OpenAI.
  • Johan Schalkwyk — former Google Fellow, early contributor to Sesame, and technical lead for Maya.
  • Pei Sun — post-training, coding, and reasoning for Gemini at Google Deepmind. Previously created the last two generations of Waymo’s perception models.
  • Jiahui Yu — co-creator of o3, o4-mini, GPT-4.1 and GPT-4o. Previously led the perception team at OpenAI, and co-led multimodal at Gemini.
  • Shengjia Zhao — co-creator of ChatGPT, GPT-4, all mini models, 4.1 and o3. Previously led synthetic data at OpenAI.

I’m excited about the progress we have planned for Llama 4.1 and 4.2. These models power Meta AI, which is used by more than 1 billion monthly actives across our apps and an increasing number of agents across Meta that help improve our products and technology. We’re committed to continuing to build out these models.

In parallel, we’re going to start research on our next generation of models to get to the frontier in the next year or so. I’ve spent the past few months meeting top folks across Meta, other AI labs, and promising startups to put together the founding group for this small talent-dense effort. We’re still forming this group and we’ll ask several people across the AI org to join this lab as well.

Meta is uniquely positioned to deliver superintelligence to the world. We have a strong business that supports building out significantly more compute than smaller labs. We have deeper experience building and growing products that reach billions of people. We are pioneering and leading the AI glasses and wearables category that is growing very quickly. And our company structure allows us to move with vastly greater conviction and boldness. I’m optimistic that this new influx of talent and parallel approach to model development will set us up to deliver on the promise of personal superintelligence for everyone.

We have even more great people at all levels joining this effort in the coming weeks, so stay tuned. I’m excited to dive in and get to work.

***

For this story, Fortune used generative AI to help with an initial draft. An editor verified the accuracy of the information before publishing. Given the nature of AI tools, mistakes may occur.

Introducing the 2025 Fortune 500, the definitive ranking of the biggest companies in America. Explore this year’s list.



Source link

Related Topics:
Continue Reading
Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

AI Research

I asked ChatGPT to help me pack for my vacation – try this awesome AI prompt that makes planning your travel checklist stress-free

Published

1 hour ago

on

July 7, 2025

By


It’s that time of year again, when those of us in the northern hemisphere pack our sunscreen and get ready to venture to hotter climates in search of some much-needed Vitamin D.

Every year, I book a vacation, and every year I get stressed as the big day gets closer, usually forgetting to pack something essential, like a charger for my Nintendo Switch 2, or dare I say it, my passport.



Source link

Continue Reading

AI Research

Denodo Announces Plans to Further Support AI Innovation by Releasing Denodo DeepQuery, a Deep Research Capability — TradingView News

Published

2 hours ago

on

July 7, 2025

By


PALO ALTO, Calif., July 07, 2025 (GLOBE NEWSWIRE) — Denodo, a leader in data management, announced the availability of the Denodo DeepQuery capability, now as a private preview, and generally available soon, enabling generative AI (GenAI) to go beyond retrieving facts to investigating, synthesizing, and explaining its reasoning. Denodo also announced the availability of Model Context Protocol (MCP) support as part of the Denodo AI SDK.

Built to address complex, open-ended business questions, DeepQuery will leverage live access to a wide spectrum of governed enterprise data across systems, departments, and formats. Unlike traditional GenAI solutions, which rephrase existing content, DeepQuery, a deep research capability, will analyze complex, open questions and search across multiple systems and sources to deliver well-structured, explainable answers rooted in real-time information. To help users operate this new capability to better understand complex current events and situations, DeepQuery will also leverage external data sources to extend and enrich enterprise data with publicly available data, external applications, and data from trading partners.

DeepQuery, beyond what’s possible using traditional generative AI (GenAI) chat or retrieval augmented generation (RAG), will enable users to ask complex, cross-functional questions that would typically take analysts days to answer—questions like, “Why did fund outflows spike last quarter?” or “What’s driving changes in customer retention across regions?” Rather than piecing together reports and data exports, DeepQuery will connect to live, governed data across different systems, apply expert-level reasoning, and deliver answers in minutes.

Slated to be packaged with the Denodo AI SDK, which streamlines AI application development with pre-built APIs, DeepQuery is being developed as a fully extensible component of the Denodo Platform, enabling developers and AI teams to build, experiment with, and integrate deep research capabilities into their own agents, copilots, or domain-specific applications.

“With DeepQuery, Denodo is demonstrating forward-thinking in advancing the capabilities of AI,” said Stewart Bond, Research VP, Data Intelligence and Integration Software at IDC. “DeepQuery, driven by deep research advances, will deliver more accurate AI responses that will also be fully explainable.”

Large language models (LLMs), business intelligence tools, and other applications are beginning to offer deep research capabilities based on public Web data; pre-indexed, data-lakehouse-specific data; or document-based retrieval, but only Denodo is developing deep research capabilities, in the form of DeepQuery, that are grounded in enterprise data across all systems, data that is delivered in real-time, structured, and governed. These capabilities are enabled by the Denodo Platform’s logical approach to data management, supported by a strong data virtualization foundation.

Denodo DeepQuery is currently available in a private preview mode. Denodo is inviting select organizations to join its AI Accelerator Program, which offers early access to DeepQuery capabilities, as well as the opportunity to collaborate with our product team to shape the future of enterprise GenAI.

“As a Denodo partner, we’re always looking for ways to provide our clients with a competitive edge,” said Nagaraj Sastry, Senior Vice President, Data and Analytics at Encora. “Denodo DeepQuery gives us exactly that. Its ability to leverage real-time, governed enterprise data for deep, contextualized insights sets it apart. This means we can help our customers move beyond general AI queries to truly intelligent analysis, empowering them to make faster, more informed decisions and accelerating their AI journey.”

Denodo also announced support of Model Context Protocol (MCP), and an MCP Server implementation is now included in the latest version of the Denodo AI SDK. As a result, all AI agents and apps based on the Denodo AI SDK can be integrated with any MCP-compliant client, providing customers with a trusted data foundation for their agentic AI ecosystems based on open standards.

“AI’s true potential in the enterprise lies not just in generating responses, but in understanding the full context behind them,” said Angel Viña, CEO and Founder of Denodo. “With DeepQuery, we’re unlocking that potential by combining generative AI with real-time, governed access to the entire corporate data ecosystem, no matter where that data resides. Unlike siloed solutions tied to a single store, DeepQuery leverages enriched, unified semantics across distributed sources, allowing AI to reason, explain, and act on data with unprecedented depth and accuracy.”

Additional Information

  • Denodo Platform: What’s New
  • Blog Post: Smarter AI Starts Here: Why DeepQuery Is the Next Step in GenAI Maturity
  • Demo: Watch a short video of this capability in action.

About Denodo

Denodo is a leader in data management. The award-winning Denodo Platform is the leading logical data management platform for transforming data into trustworthy insights and outcomes for all data-related initiatives across the enterprise, including AI and self-service. Denodo’s customers in all industries all over the world have delivered trusted AI-ready and business-ready data in a third of the time and with 10x better performance than with lakehouses and other mainstream data platforms alone. For more information, visit denodo.com.

Media Contacts

pr@denodo.com



Source link

Continue Reading

AI Research

Sakana AI: Think LLM dream teams, not single models

Published

3 hours ago

on

July 7, 2025

By


Enterprises may want to start thinking of large language models (LLMs) as ensemble casts that can combine knowledge and reasoning to complete tasks, according to Japanese AI lab Sakana AI.

Sakana AI in a research paper outlined a method called Multi-LLM AB-MCTS (Adaptive Branching Monte Carlo Tree Search) that uses a collection of LLMs to cooperate, perform trial-and-error and leverage strengths to solve complex problems.

In a post, Sakana AI said:

“Frontier AI models like ChatGPT, Gemini, Grok, and DeepSeek are evolving at a breathtaking pace amidst fierce competition. However, no matter how advanced they become, each model retains its own individuality stemming from its unique training data and methods. We see these biases and varied aptitudes not as limitations, but as precious resources for creating collective intelligence. Just as a dream team of diverse human experts tackles complex problems, AIs should also collaborate by bringing their unique strengths to the table.”

Sakana AI said AB-MCTS is a method for inference-time scaling to enable frontier AIs to cooperate and revisit problems and solutions. Sakana AI released the algorithm as an open source framework called TreeQuest, which has a flexible API that allows users to use AB-MCTS for tasks with multiple LLMs and custom scoring.

What’s interesting is that Sakana AI gets out of that zero-sum LLM argument. The companies behind LLM training would like you to think there’s one model to rule them all. And you’d do the same if you were spending so much on training models and wanted to lock in customers for scale and returns.

Sakana AI’s deceptively simple solution can only come from a company that’s not trying to play LLM leapfrog every few minutes. The power of AI is in the ability to maximize the potential of each LLM. Sakana AI said:

“We saw examples where problems that were unsolvable by any single LLM were solved by combining multiple LLMs. This went beyond simply assigning the best LLM to each problem. In (an) example, even though the solution initially generated by o4-mini was incorrect, DeepSeek-R1-0528 and Gemini-2.5-Pro were able to use it as a hint to arrive at the correct solution in the next step. This demonstrates that Multi-LLM AB-MCTS can flexibly combine frontier models to solve previously unsolvable problems, pushing the limits of what is achievable by using LLMs as a collective intelligence.”

A few thoughts:

  • Sakana AI’s research and move to emphasize collective intelligence over on LLM and stack is critical to enterprises that need to create architectures that don’t lock them into one provider.
  • AB-MCTS could play into what agentic AI needs to become to be effective and complement emerging standards such as Model Context Protocol (MCP) and Agent2Agent.
  • If combining multiple models to solve problems becomes frictionless, the costs will plunge. Will you need to pay up for OpenAI when you can leverage LLMs like DeepSeek combined with Gemini and a few others? 
  • Enterprises may want to start thinking about how to build decision engines instead of an overall AI stack. 
  • We could see a scenario where a collective of LLMs achieves superintelligence before any one model or provider. If that scenario plays out, can LLM giants maintain valuations?
  • The value in AI may not be in the infrastructure or foundational models in the long run, but the architecture and approaches.

More:



Source link

Continue Reading

Trending