☀️ AI Morning Minute: Yann LeCun
He helped invent deep learning. Now he thinks the thing he helped invent is a dead end.
Yann LeCun shared the 2018 Turing Award with Geoffrey Hinton and Yoshua Bengio for pioneering the neural network research that made modern AI possible. He spent a decade as Meta’s Chief AI Scientist. And he has spent the last several years arguing, loudly and in public, that large language models, the technology powering ChatGPT, Claude, and Gemini, will never produce human-level intelligence. That’s not a fringe position from an outsider. That’s the opinion of one of the three people most responsible for building the thing he’s criticizing.
Who they are
LeCun was born in 1960 near Paris, earned a PhD in computer science in France, and joined Bell Labs in 1988 where he developed convolutional neural networks, the architecture behind most computer vision systems still in use today. He spent years at NYU before joining Facebook in 2013 as the founding director of FAIR, the company’s AI research lab. He served as Meta’s Chief AI Scientist for over a decade, overseeing the research that produced PyTorch and the Llama model family, before leaving in November 2025 to found Advanced Machine Intelligence Labs, known as AMI Labs. The company, headquartered in Paris and focused on world models, raised $1.03 billion in March 2026 at a $3.5 billion valuation.
His departure from Meta was not quiet. He publicly called Meta’s new Chief AI Officer Alexandr Wang “inexperienced,” said Mark Zuckerberg had “basically sidelined the entire Gen AI organization,” and warned that former colleagues who hadn’t left yet soon would.
Why they matter
His technical position is a meaningful minority view with serious credentials behind it. LeCun argues that a system trained only on language will never approximate human intelligence because language is too thin a signal. Humans learn by observing and interacting with the physical world, not by reading. His focus on world models, AI systems that learn from video, sensor data, and physical dynamics rather than text, represents a genuine research bet against the current consensus.
He is the most prominent advocate for open-source AI at the frontier. Throughout his time at Meta he pushed for Llama’s open release and has consistently argued that open research accelerates progress. His position puts him in direct opposition to OpenAI and Anthropic’s closed-model approach, and he has said so in those terms, repeatedly.
The Turing Award cohort he belongs to has now split publicly on the most important question in the field. Hinton left Google to warn the world about AI risk. LeCun left Meta to argue that current AI isn’t actually dangerous because it isn’t actually intelligent. Bengio has moved toward safety advocacy. Three researchers who built the same foundation have reached three different conclusions about what it means.
What they’ve said or done
In an August 2022 paper co-authored with Jacob Browning, LeCun wrote: “A system trained on language alone will never approximate human intelligence, even if trained from now until the heat death of the universe.”
He has not softened that position since.