Meta’s departing chief AI scientist Yann LeCun isn’t holding back. In a new interview, he talks about manipulated benchmarks, a furious Zuckerberg, and a wave of departures from the AI division.<\/p>\n
Yann LeCun, one of the most influential figures in AI research, just unloaded on Meta in an interview with the Financial Times<\/a>. The 65-year-old is leaving after more than a decade to launch his own AI startup, and he’s painting a grim picture of the company’s AI operation.<\/p>\n One issue is Llama 4, Meta’s flagship language model that shipped in April 2025<\/a>. LeCun admits the published benchmarks were misleading. “Results were fudged a little bit,” he says. The team used different models for different benchmarks to game the numbers. This came out almost immediately after Llama 4’s release<\/a>.<\/p>\n CEO Mark Zuckerberg didn’t take it well: “Mark was really upset and basically lost confidence in everyone who was involved in this. And so basically sidelined the entire GenAI organization.” LeCun adds, “A lot of people have left, a lot of people who haven’t yet left will leave.”<\/p>\n LeCun recently said he’s had almost nothing to do with the LLM project since the original Llama shipped<\/a>. Meta is developing two new generative AI models codenamed “Mango” and “Avocado”<\/a> with a planned release in the first half of 2026.<\/p>\n “You certainly don’t tell a researcher like me what to do”<\/p>\n LeCun doesn’t mince words about Alexandr Wang, the 28-year-old former Scale AI CEO that Meta hired in June 2025 as part of a $14 billion deal<\/a>. Wang now runs the “TBD Lab” for frontier AI models<\/a>, which technically made him LeCun’s boss<\/a>.<\/p>\n LeCun describes Wang as “young” and “inexperienced”: “He learns fast, he knows what he doesn’t know\u2026 There’s no experience with research or how you practice research, how you do it. Or what would be attractive or repulsive to a researcher.”<\/p>\n On the new reporting structure, LeCun is direct: “You don’t tell a researcher what to do. You certainly don’t tell a researcher like me what to do.” As for Meta’s approach of luring top researchers from rivals with offers as high as $100 million<\/a>, he’s not convinced: “The future will say whether that was a good idea or not.”<\/p>\n LeCun thinks LLMs are going nowhere<\/p>\n The rift between LeCun and Meta goes beyond office politics. He sees large language models as fundamentally flawed, a “dead end” on the road to superintelligence<\/a>. But Meta’s new AI recruits are, in his words, “completely LLM-pilled,” all in on the technology he thinks is a mistake<\/a>.<\/p>\n “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,” LeCun says. “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.”<\/p>\n The disconnect between research and product teams started well before Llama 4 flopped. LeCun says they 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. When you do that, you fall behind, he adds. Eventually, staying at Meta became “politically difficult.”<\/p>\n LeCun’s new venture bets on world models<\/p>\n His new company, Advanced Machine Intelligence Labs, is run by CEO Alex LeBrun. LeCun took the Executive Chair role instead. He says he’s “too disorganized for this, and also too old” for the day-to-day stuff. Instead, his job is to inspire.<\/p>\n The startup is betting on an alternative to LLMs built around V-JEPA<\/a>, an architecture LeCun created at Meta<\/a>. Instead of learning from text alone, these models train on video and spatial data to understand how the physical world works. They can plan, reason, and hold onto information over time. LeCun calls this “Advanced Machine Intelligence” (AMI), the same name as his company. He tried to push the term<\/a> while still at Meta.<\/p>\n Early “baby” versions should be ready within a year, with full-scale systems coming a few years after that. “Maybe there is an obstacle we\u2019re not seeing yet, but at least there is hope.”<\/p>\n His new company has a global footprint but strong ties to France. French President Emmanuel Macron even texted LeCun after the news broke, though LeCun won’t say what the message said.<\/p>\n \t\t\t\tAI News Without the Hype \u2013 Curated by Humans<\/p>\n \n\t\t\t\t\tAs a THE DECODER subscriber, you get ad-free reading, our weekly AI newsletter, the exclusive “AI Radar” Frontier Report 6\u00d7 per year, access to comments, and our complete archive.\t\t\t\t<\/p>\n