Yann LeCun isn't doing the polite exit interview thing. In a Financial Times interview published this week, Meta's outgoing chief AI scientist confirmed what the AI community suspected for months: Llama 4's benchmark results were manipulated before the April 2025 release. He also took direct aim at Alexandr Wang, the 28-year-old Scale AI founder Zuckerberg paid $14 billion to bring in and, briefly, put in charge of LeCun.
The benchmark mess
LeCun didn't soften it. "Results were fudged a little bit," he told the FT. The team used different versions of the model for different benchmarks to make the numbers look better. Which is, you know, exactly what everyone accused Meta of doing when Llama 4 dropped and immediately disappointed developers who actually tried using it.
The fallout was severe. "Mark was really upset and basically lost confidence in everyone who was involved in this," LeCun said. "And so basically sidelined the entire GenAI organization."
Meta didn't announce or release a follow-up model after Llama 4's botched release. Instead, Zuckerberg went shopping. He bought Scale AI and hired Wang to run a new superintelligence lab. LeCun got shuffled into reporting to a guy half his age who'd never actually built an AI model.
"You don't tell a researcher like me what to do"
LeCun's assessment of Wang is brutal. "He learns fast, he knows what he doesn't know… 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."
The reporting structure didn't last long, obviously. But LeCun made clear it was never real: "You don't tell a researcher what to do. You certainly don't tell a researcher like me what to do."
Wang's company, Scale AI, is a data-labeling operation. They prepare training data for AI models. They don't build models. Now he's supposed to lead Meta to superintelligence. Meta's paying some of these new hires north of $100 million. One researcher from Apple is reportedly getting $200 million over four years.
More departures coming
LeCun predicts the exodus will continue. "A lot of people have left, a lot of people who haven't yet left will leave."
The relationship between Meta and Scale AI is already fraying. At least one executive Wang brought over has already departed after just two months. Meta's TBD Labs is working with competitor data-labeling vendors like Mercor and Surge, which is strange when you've just invested $14 billion in Scale AI.
The dysfunction goes deeper. Wang has reportedly told associates he finds Zuckerberg's micromanagement suffocating. Some staff question whether the 28-year-old is out of his depth.
LeCun's going to do his own thing
He's launching Advanced Machine Intelligence Labs, with Alex LeBrun (formerly of medical AI startup Nabla) as CEO. LeCun took the executive chairman title. "I'm too disorganised for this, and also too old" for day-to-day operations, he said. His job is to inspire.
The company is reportedly seeking €500 million at a €3 billion valuation before even launching a product. The bet is on "world models" built around V-JEPA, an architecture LeCun developed at Meta. Instead of learning from text, these models train on video and spatial data to understand physics and plan actions.
He's been saying for years that LLMs are a dead end for real intelligence. Meta's new AI team is, in his words, "completely LLM-pilled."
"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 said. "But I'm not gonna change my mind because some dude thinks I'm wrong. I'm not wrong."
What comes next
Meta's secretive TBD lab is aiming to release an entirely new AI model built from scratch in the first quarter of 2026. LeCun says early versions of his world model approach should be ready within a year, full-scale systems a few years after that.
French President Emmanuel Macron texted LeCun after the news broke. LeCun won't say what the message said.
