It’s news that has shaken the AI community: Yann LeCun, one of the pioneers of deep learning and an iconic figure in Meta’s research for over ten years, has decided to quit.
But the Turing Award laureate isn’t retiring. He wants to create his own independent lab to pursue a vision of AI that he deems incompatible with Mark Zuckerberg’s company’s current priorities.
Business AI and scientific research: incompatible?
For years, Yann LeCun has been repeating the same thing, with almost provocative consistency: large language models (LLMs), which everyone is scrambling for, will never lead to truly human intelligence.
They can produce fluent, sometimes astonishing sentences, but they don’t understand anything they’re saying. No persistent memory, no structured reasoning, no internal model of the physical world.
For Yann LeCun, these systems merely identify statistical correlations in mountains of text. They predict the most probable next part of a sentence, full stop.
Instead, he has long advocated for the idea of “World Models“: architectures that would learn like children or animals, through observation, predicting the consequences of their actions, and progressively building a coherent representation of reality. A much slower, much more uncertain path… and far less profitable in the short term.
Meta chooses acceleration in super AI
Meanwhile, Meta has taken a radical turn. The company has just created Superintelligence Labs, led by Alexandr Wang, the young CEO of Scale AI. The stated goal: to catch up with OpenAI and Google at full speed.
LLaMA 4 didn’t convince, Meta AI chatbot adoption remains anemic, and shareholders are growing impatient. After pouring over $100 billion into AI, Meta even cut 600 positions in its research teams last fall.
The message is clear: we need products that generate revenue, fast.
Yann LeCun’s voice, advocating for patience and conceptual breakthroughs, therefore became difficult to hear. Increasing model size, stacking GPUs, scaling again and again: this is the dominant religion of Silicon Valley, but not that of the French researcher.
A broader symptom
This departure is not isolated. Geoffrey Hinton, the other great Turing Award laureate and inventor of the foundations of modern deep learning, left Google two years ago. Different reasons (existential risks of AI) but the same unease: the impression that the race for raw power has taken precedence over reflection.
The same tension was found at OpenAI, where Elon Musk and Sam Altman parted ways over a philosophical question: maintain a non-profit structure dedicated to safety, or raise tens of billions from Microsoft to dominate the market? Sam Altman made his choice, and Elon Musk left to create xAI.
Towards a two-speed AI?
Yann LeCun’s departure may mark the beginning of a lasting split. On one side, the giants betting everything on scaling and immediate applications: chatbots, assistants, image generation, targeted advertising. Facing them, a handful of researchers who refuse to sacrifice science on the altar of stock prices.
By creating his own structure, Yann LeCun regains the freedom to think long-term, far from Excel spreadsheets and quarterly announcements.
It remains to be seen whether this parallel path will manage to catch up with the billions invested by the GAFAM companies… or if it will ultimately prove that we’ve all been on the wrong track from the start.
But when two Turing Award laureates successively leave industry giants, it’s not just a job change. It’s a warning sign. AI is moving at full speed, but no one really seems to know where it’s going!
Source: Financial Times
