Yann LeCun, Meta’s chief AI scientist and a Turing Award laureate, is reportedly planning to leave the company to start his own AI-focused startup. According to the Financial Times, LeCun intends to pursue research on a different kind of artificial intelligence known as “world models,” which contrasts with the large language model (LLM) approach that Meta and many other tech giants have recently prioritized. This departure signals a significant shift within Meta, especially amid CEO Mark Zuckerberg’s recent overhaul of the company’s AI strategy—an overhaul driven by a sense that Meta has fallen behind competitors such as OpenAI and Google.
LeCun, a French-American scientist renowned for his foundational contributions to deep learning and convolutional neural networks, founded Meta’s Fundamental AI Research lab (FAIR) in 2013 and has served as the company’s chief AI scientist ever since. Despite his long tenure and central role in shaping Meta’s AI research, he has reportedly expressed frustration with the company’s pivot away from fundamental research toward rapid product development and commercialization. His planned exit, which insiders say will take place over the coming months, is already accompanied by early fundraising talks for his new venture.
The startup will focus on “world models,” a concept in AI research that aims to build systems capable of developing an internal understanding of the physical world. Unlike current large language models, which primarily predict the next word or token in a sequence of text data, world models are designed to simulate cause-and-effect relationships, understand physical laws, and enable machines to reason, plan, and navigate environments in a way more akin to how animals do. This approach relies heavily on learning from video and spatial data rather than text alone, potentially allowing AI to grasp the dynamics of the physical world more intuitively. LeCun has previously suggested that such models could take a decade or more to fully mature, emphasizing the long-term nature of this research.
The distinction between current AI architectures and world models is important. While some experts argue that transformer-based models—such as those powering ChatGPT and other large language models, video synthesis tools, and interactive world synthesis systems—have shown emergent abilities suggestive of understanding physical principles, the consensus remains that these models fundamentally operate through sophisticated pattern recognition. They excel at identifying and replicating patterns in data but lack a deep, causal understanding of how the world works. Meta’s shift toward large language models, therefore, represents a focus on technologies that can be quickly deployed to market but might not fundamentally advance the deeper AI capabilities that LeCun envisions.
LeCun’s departure comes amid a turbulent year for Meta’s AI division. Earlier in 2024, the company faced criticism over the launch of Llama 4, its AI language model, which many industry observers considered underwhelming compared to offerings from competitors like Google, OpenAI, and Anthropic. The launch was marred by accusations of benchmark gaming and disappointing performance results, which damaged Meta’s reputation in the AI community. Additionally, Meta’s AI chatbot has struggled to gain a foothold with consumers and has been embroiled in controversies, particularly around its interactions with children, further complicating the company’s AI ambitions.
Within Meta, the tension between long-term AI research and immediate product development has been growing. LeCun’s lab, FAIR, has traditionally focused on fundamental AI research, but recent organizational changes reflect a strategic pivot. Over the summer of 2024, Mark Zuckerberg made significant moves to accelerate product-focused AI development by hiring Alexandr Wang to lead a new “superintelligence” team. Wang, a 28-year-old founder of the data-labeling startup Scale AI, was brought onboard through a $14.3 billion deal that included acquiring a 49 percent stake in his company. This restructuring placed LeCun under Wang’s leadership, a move perceived by some as a rebuke of LeCun’s more research-oriented approach.
At the same time, Zuckerberg assembled an elite team known as the TBD Lab, tasked with rapidly advancing the next generation of large language models. To attract top talent, Meta offered extraordinarily generous compensation packages, reportedly ranging from $100 million to $250 million, drawing experts away from rival firms like OpenAI and Google. This aggressive hiring and investment spree underscores Zuckerberg’s determination to catch up in the AI race and deliver commercial products that will drive Meta’s growth.
However, this strategy has been met with skepticism both inside and outside the company. LeCun himself has