Why Did Facebook's AI Pioneer LeCun Leave for a Risky Startup? The Shocking Truth Revealed!

Yann LeCun, the departing chief AI scientist at Meta, has candidly expressed his concerns about the company's direction and internal turmoil in a recent interview with the Financial Times. After more than a decade at Meta, LeCun is leaving to establish his own AI startup, painting a troubling picture of the AI division he is leaving behind.

LeCun's criticisms come in the wake of the release of Meta's flagship language model, Llama 4, in April 2025. He admits that the benchmarks published to showcase the model's performance were "manipulated" to present favorable results. "Results were fudged a little bit," he stated, indicating that different models were employed for different benchmarks to artificially enhance the numbers. This controversy emerged soon after Llama 4's debut, leading to a significant fallout within the company.

Mark Zuckerberg, Meta's CEO, reportedly reacted with anger to the revelations, losing confidence in those involved in the project. LeCun noted, "Mark was really upset and basically lost confidence in everyone who was involved in this. And so basically sidelined the entire GenAI organization." He warned that many talented individuals have already left and more are expected to follow.

Since the launch of the original Llama model, LeCun claims he has had limited involvement with the project. Meanwhile, Meta is working on two new generative AI models, codenamed "Mango" and "Avocado," which are slated for release in the first half of 2026.

Internal Struggles at Meta

The internal dynamics at Meta's AI division have also come under scrutiny. LeCun has been critical of Alexandr Wang, the 28-year-old former CEO of Scale AI, who was brought on board in June 2025 as part of a $14 billion acquisition. Wang now oversees the "TBD Lab," which focuses on frontier AI models and has effectively become LeCun's superior. LeCun described Wang as "young" and "inexperienced," adding, "There’s no experience with research or how you practice research." He firmly believes that researchers should not be directed in their work, stating, "You don't tell a researcher what to do. You certainly don't tell a researcher like me what to do."

LeCun is skeptical about Meta's strategy of attracting top talent from rivals with exorbitant offers, some reaching as high as $100 million, expressing uncertainty over its long-term effectiveness: "The future will say whether that was a good idea or not."

Beyond office politics, LeCun has profound concerns about the very nature of large language models (LLMs). He regards them as fundamentally flawed and considers the current focus on LLMs a "dead end" on the path to achieving superintelligence. "I'm sure there's a lot of people at Meta, including perhaps Alex, who would like me to not tell the world that LLMs basically are a dead end when it comes to superintelligence," he remarked. "But I'm not gonna change my mind because some dude thinks I'm wrong. I'm not wrong. My integrity as a scientist cannot allow me to do this."

LeCun believes that the disconnect between research and product teams was evident long before the issues surrounding Llama 4 surfaced. "We had many new ideas and 'really cool stuff' that the product teams should have implemented, but they just went for things that were essentially safe and proven," he explained. He feels that this approach has led to a stagnation that ultimately made his role at Meta politically untenable.

With his new venture, Advanced Machine Intelligence Labs, LeCun is pivoting away from LLMs. The startup is led by CEO Alex LeBrun, while LeCun takes on the role of Executive Chair. He describes himself as "too disorganized for this, and also too old" for day-to-day operations, preferring to inspire the team instead. The company aims to develop an alternative architecture called V-JEPA, which leverages video and spatial data to create models capable of understanding the physical world. "These models can plan, reason, and hold onto information over time," he explained. He refers to this approach as "Advanced Machine Intelligence" (AMI), the same name as his new company.

LeCun anticipates that early versions of these models could be ready within a year, with fully operational systems to follow in a few years. "Maybe there is an obstacle we’re not seeing yet, but at least there is hope," he expressed. His new company not only has a global reach but maintains strong connections to France, with French President Emmanuel Macron even reaching out to LeCun following the announcement of his new venture.