According to Business Insider, Yann LeCun, Meta’s chief AI scientist for 12 years, is leaving the company to start his own AI startup focused on world-model research. Meta confirmed the departure and revealed they’ll partner with LeCun on his new venture, though details about the partnership remain unclear. The move comes during a period of instability within Meta’s AI organization, which has seen dozens of new hires from rivals and a major August reorganization creating four distinct AI teams. This follows Meta’s pivot toward competing with OpenAI, Google DeepMind, and Anthropic on large-scale AI models, while researchers have cycled in and out of key roles. Earlier this week, Soumith Chintala, creator of Meta’s PyTorch framework, also left after 11 years to join Mira Murati’s Thinking Machines Lab.
Meta’s AI shakeup continues
Here’s the thing: LeCun’s departure isn’t happening in isolation. Meta’s AI division has been undergoing what looks like a complete transformation. They brought in former Scale AI CEO Alexandr Wang to lead their new Superintelligence Labs division, hired dozens of top researchers from competitors, and completely reorganized their AI operations in August. But apparently that’s created some serious internal friction. Business Insider previously reported tensions between the highly paid new hires and existing researchers, with some veterans threatening to quit. So LeCun walking away after 12 years? That’s a pretty significant vote of no confidence in the new direction.
A philosophical divide on AI’s future
What’s really interesting here is the fundamental disagreement about where AI should be heading. LeCun has been openly critical of the industry‘s obsession with large language models. He’s been pushing his JEPA approach – which focuses on training AI to understand the physical world from sensory data rather than just generating text. Meanwhile, Meta has been going all-in on scaling LLMs and commercially driven model development. Basically, LeCun wants to build AI that understands how the world works, while Meta wants to build AI that can write better marketing copy. Can you blame him for wanting to pursue his vision elsewhere?
What this means for Meta and AI
Losing someone like LeCun is a massive blow to Meta’s AI credibility. He’s not just another researcher – he’s one of the “godfathers of AI” and has been the public face of Meta’s AI research for over a decade. His departure, combined with Chintala leaving PyTorch, suggests there’s a brain drain happening at the exact moment Meta needs its best minds to compete with OpenAI and Google. And let’s be honest – Meta’s recent Llama 4 release got pretty muted reactions both inside and outside the company. Now they’re losing the very people who built their AI foundation. That doesn’t exactly inspire confidence in their ability to catch up in the AI race.
startup-opportunity”>The startup opportunity
The fact that Meta is partnering with LeCun on his new venture is fascinating. It suggests they recognize his value even as he walks out the door. World-model research is exactly the kind of foundational work that could lead to the next breakthrough in AI – the kind that might eventually make today’s LLMs look primitive. If LeCun can build AI that truly understands cause and effect in the physical world, that’s potentially huge for everything from robotics to autonomous systems. And honestly? Starting fresh without the corporate baggage of Meta might be exactly what he needs to pursue that vision without compromise.
