A major shake up appears to be taking place at Meta.
According to new reporting from the Financial Times, the company’s top AI scientist, Yann LeCun, is planning to leave his position there and start his own AI startup.
It’s a genuinely startling decision. LeCun, who joined Meta in 2013, is a towering figure in the AI industry. A Turing Award winner, the 65-year-old is considered to be one of the three so-called “godfathers” of modern AI for his pioneering work on neural networks, the tech that underpins the large language models used by much of the AI industry. As such, his presence at Meta has imbued the company’s often-struggling AI efforts with an aura of credibility.
LeCun’s departure comes as CEO Mark Zuckerberg has carried out sweeping changes to Meta’s approach to AI. Whereas before the company had a heavy focus on research and open source models, it’s now retooling its efforts to focus more on fielding commercially competitive AI products, as its AI chatbots lags behind rivals, like Google and OpenAI.
LeCun is famously something of an LLM skeptic, believing that the architecture is incapable of one day achieving human-levels of cognition, resulting in a so-called artificial general intelligence. He even advised up-and-coming programmers to not pursue LLMs at all, and instead work “on next-gen AI systems that lift the limitations of LLMs.” That makes him an outlier in the industry, as one of the driving promises fueling the boom is that the tech provides a direct line to creating AGI, if it isn’t already on the verge of doing so. With his focus on more esoteric forms of AI and his distaste of AI boosterism, LeCun was always an odd figure to be working at a titan like Meta.
Perhaps these contradictions finally became irreconcilable. This summer, Meta poured over $14 billion in the AI data annotation startup Scale AI and poached its then-CEO Alexandr Wang to lead a newly created Superintelligence Labs. Separate from LeCun’s research division, FAIR, it aims to create a “superintelligent” AI using LLM technology. LeCun, by contrast, is adamant about creating “world” models that are designed to understand the three-dimensional world by training them on a variety of physical data, rather than only language. LeCun says these advances could take decades, but Zuckerberg is clearly obsessed with market dominance in the immediate term.
It’s also evident now which of the AI approaches are the favorite child. Zuckerberg has tried to bring in talent to the superintelligence division by offering astronomical contracts worth hundreds of millions of dollars. LeCun was also forced to start reporting to the 28-year-old Wang once he joined Meta, the FT noted.
The change of gears to developing “superintelligence” came after Meta’s newest Llama 4 model was widely considered a disappointment with its lackluster performance compared to competitors’ efforts.
The market, however, hasn’t been fully confident with Zuckerberg’s new direction. Last month, its stock plunged by more than 11 percent following a third quarter earnings call in which it projected spending billions of dollars more on AI this year than earlier guidance suggested. Following the report of LeCun’s planned departure, Meta’s stock dipped by nearly another 3 percent.
More on Meta: Meta Accuses Employee’s Dad of Downloading Gigantic Illegal Goon Stash