In a move that’s sending shockwaves through the tech world, Yann LeCun, the renowned “Godfather of AI” and Meta’s Chief AI Scientist, has announced his departure on Nov 19, 2025 from the company after over a decade of groundbreaking work. At 65, LeCun isn’t retiring, he’s launching a new startup aimed at revolutionizing artificial intelligence with a focus on “Advanced Machine Intelligence” (AMI), a paradigm that promises to outstrip today’s language models by building systems that truly understand the physical world.

As of today, December 31, 2025, LeCun’s exit from Meta is now complete, marking the official start of his next chapter, one that could redefine how AI understands and interacts with the physical world.

Reflections on a Transformative Tenure

LeCun’s journey at Meta began in 2013, where he founded and led the Fundamental AI Research (FAIR) lab, contributing to pivotal advancements in deep learning, convolutional neural networks, and open-source AI tools like PyTorch. His tenure wasn’t without controversy; LeCun has been a vocal critic of doomsday AI narratives, famously clashing with figures like Elon Musk over the risks of superintelligence. But as Meta shifted toward product-driven AI like Llama models and metaverse ambitions, rumors swirled that LeCun felt sidelined. In a LinkedIn post confirming his exit, he reflected on his 12 years, hinting at a desire to pursue unconstrained research.

The Game-Changing VLJ Breakthrough

The timing couldn’t be more intriguing. Just before his announcement in November 2025, LeCun dropped a bombshell paper on VLJ (Vision-Language Joint Embedding Predictive Architecture), an extension of Meta’s V-JEPA model. Unlike traditional large language models (LLMs) such as GPT-4 or Grok, which generate text token by token, VLJ operates in a “semantic space” predicting meanings directly rather than words. This non-generative approach makes it faster, more efficient (using about half the parameters of rivals), and better suited for real-world applications like robotics and autonomous agents.

How VLJ Achieves Real-World Understanding

Imagine an AI that doesn’t just chat but comprehends videos, images, and actions in a continuous stream, tracking changes over time with a built-in memory. In demos, VLJ stabilizes insights across frames starting with initial guesses (red dots) that solidify into confirmed understanding (blue dots) far outperforming cheaper models that inconsistently describe scenes without context. LeCun argues that true intelligence isn’t about mastering language but understanding the world’s causal dynamics, enabling “world models” for planning, reasoning, and even counterfactual thinking (like “what if?”). This could transform fields from self-driving cars to wearable tech, where pixel-level noise is abstracted away in latent spaces.

The Vision for a New Startup

Sources confirm that LeCun’s new venture is expected to focus on developing “world models,” advanced AI systems designed to understand and reason about the physical world rather than relying solely on text-based learning. The startup, likely based in Paris and called Advanced Machine Intelligence Labs, is already in early fundraising talks for around €500 million, reflecting strong investor confidence in his vision. Analysts predict the company could reach a multibillion-euro valuation before officially launching in early 2026.

Implications for Meta and the AI Industry

For Meta, losing LeCun marks the end of an era, though the company insists its AI momentum remains strong under CEO Mark Zuckerberg’s vision. Yet, this exit underscores a broader industry shift: from hype-driven chatbots to practical, embodied intelligence.

A Philosophical Pivot for the Future

As LeCun steps into entrepreneurship, the AI community watches eagerly could VLJ and AMI herald the “new kind of LLM” that finally bridges the gap between digital brains and the real world? This development isn’t just a personnel change; it’s a philosophical pivot. In a field dominated by generative giants, LeCun’s bet on predictive, world-modeling AI might just redefine what’s possible. Stay tuned 2026 could be the year AI gets real.