Yann LeCun, who spent more than a decade as Meta’s chief AI scientist and is widely credited as one of the foundational researchers behind modern deep learning, is departing the company to launch a new AI research startup called Advanced Machine Intelligence Labs. LeCun will take the role of Executive Chairman at the new venture, with Alex LeBrun serving as CEO — a structure that lets LeCun focus on research direction and vision while day-to-day operational leadership sits with someone else, a governance pattern echoed elsewhere in the AI industry this year, including at Anduril and Microsoft AI.
The new startup’s stated research focus is deliberately different from the large language models that have dominated AI progress and funding for the past several years: LeCun has said Advanced Machine Intelligence Labs will build “world models” — AI systems trained on images, video, and spatial data rather than relying primarily on text, aimed at giving AI systems genuine understanding of physical reality, persistent memory, and the ability to reason through and plan complex, multi-step real-world action sequences. LeCun has been publicly skeptical for years that scaling up text-based large language models alone will ever produce AI with real physical-world understanding, making this new venture a direct, well-funded test of that long-standing public position rather than a new idea he’s only just begun advocating for.
A Clean Departure, Not a Rupture
Notably, LeCun’s exit from Meta was not framed by either side as an acrimonious split. Meta has agreed to partner with the new startup, with LeCun stating publicly that some of the research will overlap with Meta’s own commercial AI interests and some of it will pursue directions Meta itself has no direct stake in — a genuinely collaborative arrangement rather than a competitor cleanly severing all ties with a former employer. LeCun is reportedly in early discussions to raise roughly €500 million to fund the new venture, a substantial sum reflecting investors’ confidence in his research reputation even before the new company has shipped any product.
A Long-Standing, Public Scientific Disagreement
LeCun’s departure gives him a well-resourced platform to test a scientific position he’s held publicly, and often controversially, for years: that today’s dominant large language model architectures, however impressive their text-generation capabilities, are fundamentally the wrong foundation for AI systems that need genuine understanding of physical cause and effect, spatial reasoning, or long-term planning in the real world. That view has put him in occasional public tension with other prominent AI researchers and executives who see continued scaling of existing large language model approaches as the more promising near-term path — a genuine, unresolved scientific and commercial debate playing out among the people actually building these systems, not a settled question either side can claim to have already won. LeCun’s own standing as one of the field’s most cited and credentialed researchers — a recipient, alongside Geoffrey Hinton and Yoshua Bengio, of the 2018 Turing Award for pioneering work in deep learning — means his specific choice to bet a new company on world models rather than large language models carries real, independent scientific weight beyond just another well-funded AI startup pursuing a differentiated pitch to investors. Whether world models eventually prove to be the missing ingredient LeCun believes they are, or whether continued scaling of existing large language model approaches closes the physical-reasoning gap on its own, is a question the wider AI research community, not just LeCun’s own new company, will be actively testing over the next several years, and the outcome will likely shape which research bets the next wave of well-funded AI startups choose to make. LeCun’s own decision to test the theory with a dedicated, well-capitalized company of his own, rather than simply continuing to argue the point from within a large existing lab, is itself a meaningful escalation of how seriously he takes the disagreement — a real, near-term opportunity for outside observers to see whether the theory holds up once it has to produce something more concrete than an argument.
What This Means for Philippine Founders
LeCun’s departure is a genuine, current signal that not every leading AI researcher believes the industry’s current, near-total focus on large language models is the only viable path forward — a useful reminder for Philippine AI startups to periodically reassess whether their own product roadmap is chasing the most crowded, well-funded category by default, versus a genuinely differentiated technical bet that a credible researcher is willing to leave a stable, well-resourced position to pursue. The LeCun-Meta partnership structure — a departing researcher’s new venture explicitly collaborating with, rather than purely competing against, their former employer — is also a useful model for any Philippine founder navigating an amicable exit from a larger company: a clean, non-adversarial separation that preserves a working relationship can create real, ongoing commercial value for both parties that a more contentious departure would foreclose entirely.
Share this article