About Yann LeCun
Yann LeCun is the Chief AI Scientist at Meta and a Turing Award winner (2018, with Geoffrey Hinton and Yoshua Bengio). He pioneered convolutional neural networks in the 1980s-90s and is now one of AI's most vocal skeptics of LLM-only approaches to AGI.
Career Highlights
- Meta (2013-present): Chief AI Scientist, VP of AI Research
- AMI (2025): Co-founder of Advanced Machine Intelligence startup
- Turing Award (2018): With Hinton and Bengio for deep learning
- NYU (2003-present): Silver Professor of Computer Science
- Bell Labs (1988-2002): Invented convolutional neural networks
Notable Positions
On LLMs Being Insufficient
LeCun's contrarian thesis:
"You cannot get to human-level AI through text alone. Training an LLM requires 30 trillion tokens - effectively all internet text. That same 10^14 bytes represents just 15,000 hours of video - 30 minutes of YouTube uploads."
On World Models (JEPA)
His alternative approach:
"JEPA predicts in abstract representation space, not pixel space - eliminates unpredictable details while preserving structure for planning."
On Open Research
"You cannot call it research unless you publish - internal hype creates delusion. Scientists need external validation."
Key Quotes
- "LLMs cannot get us to human-level AI."
- "World models trained on video, not text."
- "You cannot call it research unless you publish."
Related Reading
- World Models - LeCun's path to AGI
- JEPA - Joint Embedding Predictive Architecture
Video Appearances
Data efficiency of vision vs text
You cannot get to human-level AI through text alone. Training an LLM requires 30 trillion tokens - all internet text. That same data is just 15,000 hours of video - 30 minutes of YouTube uploads.
at 00:10:00
Open research philosophy
AMI will publish openly because you cannot call it research unless you publish - internal hype creates delusion.
at 00:20:00
Joint Embedding Predictive Architecture
JEPA predicts in abstract representation space, not pixel space - eliminates unpredictable details while preserving structure.
at 00:15:00
LLM limitations and world models
Dr. Yann LeCun argues LLMs are 'an off-ramp in the highway of AI studies' - impressive but limited as token-to-token generators. He left Meta to create AMI, a startup focusing on world models.
at 01:55:00