Meta’s AI research lab is ‘dying a slow death,’ some insiders say. Meta prefers to call it ‘a new beginning’

The timing of Pineau’s resignation raised eyebrows. It came just days before an unusual weekend rollout of Meta’s Llama 4 models that wound up being surrounded by controversy. The new models drew criticism from the research community over a perceived rushed release, lack of transparency, possibly inflated performance metrics, and indications that Meta was failing to keep pace with open-source AI rivals like China’s DeepSeek. It all comes at a time of intense competition in the AI model market, with Meta planning to spend up to $65 billion this year on AI infrastructure. 

FAIR—an acronym for Fundamental AI Research—was once the crown jewel of AI development at Meta. But as Mark Zuckerberg has pivoted the company toward generative AI products over the past two years, the vaunted lab has become something of an orphan inside the organization, increasingly shoved out of the limelight by more commercially focused AI groups within the company. The newest Llama model, for instance, was the product of Meta’s separate GenAI team, not FAIR. Meanwhile, FAIR has languished, with talented researchers departing for rival companies and startups: More than half of the 14 authors of the original Llama research paper published in February 2023 had left the company six months later, while at least eight top researchers have left over the past year. 

The lab has been “dying a slow death,” according to one former Meta employee who spoke to Fortune over the past week, echoing a sentiment from other ex–FAIR team members. Pineau’s departure, in this view, can be interpreted as the lab’s death rattle.

Yann LeCun, Meta’s chief AI scientist, who is considered one of the “godfathers” of deep learning and who founded FAIR, is now leading the team temporarily as Meta does an external search to replace Pineau (who is staying on until May). He denies FAIR is fading into oblivion, telling Fortune by email that “this is most definitely not the death of FAIR.” To the contrary, he insists the lab is about to experience a new dawn, and the existence of the GenAI product group actually enables FAIR to refocus on longer-term AI research. 

“It’s more like a new beginning in which FAIR is refocusing on the ambitious and long-term goal of what we call AMI (advanced machine intelligence),” LeCun said. The 64-year-old French-born computer scientist has long argued that the term artificial general intelligence (AGI) is misleading, and coined the term AMI, which he says is about helping machines understand the world, reason, plan, and learn as efficiently as animals and humans. He also has gone on record as skeptical that current LLM-based approaches to AI will ever get to human-level intelligence. 

Pineau also sent Fortune a statement, saying that she ”continues to be very enthusiastic about Meta’s overall AI work and strategy.” There continues to be strong support for exploratory research and FAIR as a distinct organization in Meta, she said, but “the time was simply right for me personally to refocus my energy before jumping into a new adventure.”

However, Fortune spoke over the past week with seven former Meta employees, some of whom requested anonymity because they were not authorized to speak about their former employer. Many of the employees, who include ex–FAIR researchers, say the organization has been slowly but surely withering in recent years as “blue sky” research (that is, fundamental research that seeks new knowledge without necessarily focusing on immediate practical applications or commercial gains) within Big Tech companies has slowed. FAIR also gets access to less computing power for its projects than the teams focused on generative AI, they say—though Meta would not confirm or deny that, saying in a statement that it does not comment on compute allocation. 

FAIR has done foundational AI research work for over a decade

Founded in December 2013 by Zuckerberg and LeCun, the lofty mission statement for FAIR was advancing “the state of the art in artificial intelligence through open research for the benefit of all.” FAIR has done foundational work in AI research that laid the groundwork for future advances in everything from computer vision and natural language processing to robotics. Many of FAIR’s innovations did wind up helping Meta with its products, from tagging friends’ photos on social media to helping moderate content for extremism and hate speech. 

David Paul Morris—Bloomberg/Getty Images

FAIR, which now boasts around 1,000 employees according to the Financial Times, began with locations at Facebook’s headquarters in Menlo Park, Calif., London, and Manhattan, expanding in 2015 to Paris and adding smaller satellite labs in Seattle, Pittsburgh, Tel Aviv, and Montreal. 

But by 2022, Meta had folded the FAIR team into Reality Labs—the division focused on augmented and virtual reality technologies for the so-called metaverse—prompting many FAIR researchers to leave. Despite this seeming demotion, FAIR was back in the spotlight in 2023, playing a central role in what was seen as Meta’s remarkable comeback in generative AI. 

When OpenAI’s ChatGPT launched in late November 2022, Meta was perceived as having fallen far behind OpenAI, Anthropic, and Google in generative AI. But FAIR helped get Meta back in the game by developing the first freely available generative AI model that could rival the models being offered by those other companies. It was called Llama—a play on the acronym for large language models, or LLMs, the type of AI that powered innovations like ChatGPT—and it took the AI world by storm.

FAIR quickly followed up the first Llama model with Llama 2 in July 2023. Its release sparked a wave of open-source LLM development and positioned Meta as a serious contender in the LLM space. By September of that year though, trouble was afoot: The Information reported that many of the original FAIR researchers who worked on Llama had left over an internal battle over computing resources with another Meta research team working on a rival model that the company ultimately abandoned. Two of the departing researchers went on to found Mistral AI, a French startup now valued at $6 billion; Armand Joulin, a Llama author and a research director at FAIR, decamped to Apple.

A few months later, in January 2024, FAIR was once again restructured. As the generative AI boom accelerated and Meta ramped up efforts to compete with companies including OpenAI, Google, and Elon Musk’s xAI, FAIR and Meta’s generative AI product team under Ahmad Al-Dahle, called GenAI, were consolidated into a single group, which a former FAIR leader described as a “blow.”

That blow related to the work on Llama, as well: The Llama 4 release last weekend was led by Meta’s GenAI product organization, with VP of generative AI Ahmad Al-Dahle at the helm. Pineau and FAIR had developed the original Llama model and Llama 2. But after that, a team within the newly launched GenAI organization led by Meta VP Manohar Paluri took over the Llama work, leaving FAIR mostly on the sidelines. 

Ex-employees say Meta has deprioritized exploratory AI research

Since the consolidation of FAIR into the product organization, ex-employees say Meta has steadily deprioritized the kind of open-ended, exploratory research that FAIR was known for, shifting resources instead toward product-driven initiatives under GenAI.

“To be honest, it makes me really sad,” said one former FAIR researcher who left in 2023 to found a startup. “FAIR at its peak circa 2019 was a very special place.” He added that he thinks Meta CEO Mark Zuckerberg “clearly values GenAI over FAIR at this point.”   

Nathan Laine—Bloomberg/Getty Images

Another former FAIR researcher, who left in 2021, recalled that FAIR had traditionally been more open to projects that explored a wide range of AI subfields—of which generative AI was just one. “It used to be a place where you could pitch a project to your manager and just do it,” he said. “You could say, ‘I want to work in robotics,’ ‘I want to work in computer vision,’ ‘I want to work in speech recognition’ … and it was very collaborative, too.” 

Now, he said, he sees FAIR, along with the AI research labs at other companies like Microsoft and Google, becoming less supportive of an academic mindset. “This is happening industrywide,” he said. “More and more people are being forced to move into generative AI.” Google, for example, merged its two AI research organizations, DeepMind and Google Brain, into a single entity called Google DeepMind that is now more focused on helping produce Google’s Gemini AI models than pursuing foundational AI research as the two labs were previously. 

In written comments to Fortune, Meta said its commitment to FAIR remains strong. “Our strategy and plans will not change as a result of recent developments, and we remain dedicated to leading AI research,” the company wrote. “Our FAIR and GenAI teams work closely together and allow for better coordination across teams [and] faster decision-making. We’re excited about the new opportunities ahead, which includes FAIR’s continued focus on advanced machine intelligence (AMI), which will enable us to tackle ambitious long-term research goals.” 

But the former employees say FAIR has undoubtedly changed, and not for the better. Today, rather than research-focused technologies to advance the AI field, everything is channeled into building AI-powered products, one said, adding that Zuckerberg is not research-focused but focused on the promise of AI to Meta’s bottom line. 

Some former FAIR researchers say a product focus is natural

William Falcon, founder and CEO of Lightning AI, did some of his PhD research while at FAIR in 2019. While he feels nostalgic for FAIR’s storied past, he is realistic about its future as a blue-sky research lab in an era of rapid AI product development.

“You could do what you do as a professor with a lot more compute and [were] paid a lot of money,” he said. However, he also said that was essentially a “honeymoon period” and that it’s a “natural evolution” to get back to product building. 

Erik Meijer, a former Meta engineer and researcher whose team was laid off in 2022, said that he was “never a fan” of research divisions inside companies like Meta. 

“If companies want to do fundamental research, they should give money with no strings attached to universities,” he said. “Industrial labs should work closely with the product teams to create a pipeline of future innovations via a tight feedback loop between production and research.” 

To be blunt, he said, “I think it is great if FAIR disappears and that Meta is putting, to borrow a Google saying, ‘More wood behind fewer arrows.’” He was referring to former Google CEO Larry Page’s use of the expression to describe the company’s strategy of focusing its resources on a small number of core projects. 

One of the former FAIR researchers emphasized that he did not believe FAIR would disappear. 

“It’s a cycle,” he said of the current product focus across the AI industry, which he acknowledged could be frustrating for researchers. “It’s time for exploitation, but exploration will come back very soon, that’s why these companies keep their blue-sky team,” he said. “As Yann says, it’s probably the perfect timing to start fresh, so they are prepared to be on top of the next wave of innovations.” 

It remains to be seen whether FAIR can ever reclaim its former glory. For now, Meta told Fortune in a statement, chief product officer Chris Cox, LeCun, and Pineau (who remains at Meta until May) are “already working to find a successor [to lead FAIR] and look forward to continuing our work in AI research.” 

Yet, the same statement doubled down on Meta’s product focus, saying that FAIR’s long-term research goals “will ultimately allow us to build our best products, adding more comprehensive capabilities like reasoning, planning, and coding that will enable new forms of utility and bring us much closer to delivering human-level experiences.” 

No matter FAIR’s fate, the research lab’s winding trajectory somewhat mirrors that of OpenAI, which was founded as a nonprofit research lab in 2015, and is now working to transition to a for-profit company after seeing ChatGPT become the fastest-growing consumer application in history since its launch in 2022. 

Of course, the app that held that title prior to ChatGPT was … Facebook. Zuckerberg would love to reclaim that mantle. It’s one of the reasons he’s laser-focused on building AI products for today’s commercial market and why FAIR will continue to play second fiddle to Meta’s GenAI division. One shouldn’t count LeCun out, of course. As long as he remains at Meta—and if he’s right that human-level AI will require a fundamentally different approach than today’s LLMs—then FAIR might just be the lab that delivers that breakthrough. For now, however, FAIR’s glory days are over.