The Information Bottleneck EP20: Yann LeCun
20 snips
Dec 15, 2025 
Yann LeCun, a Turing Award-winning computer scientist and pioneer of deep learning, shares his bold vision for AI after leaving Meta to start Advanced Machine Intelligence. He critiques the current Silicon Valley obsession with scaling language models, arguing they won't lead to artificial general intelligence. Instead, he advocates for developing world models that simulate abstract concepts. Yann discusses learning object permanence and the challenges of game AI, while advocating for safety measures in AI design, emphasizing an architecture-focused approach.
AI Snips
Chapters
Transcript
Episode notes
Abstract World Models Over Pixel Simulators
- World models should predict consequences in abstract representation space instead of pixels.
- Planning uses those predictions to optimize action sequences for tasks.
Publish Upstream Research
- Publish upstream research to avoid self-delusion and improve scientific rigor.
- Open publication motivates researchers and enables long-term breakthroughs.
Why Representation Prediction Beats Pixel Prediction
- Predicting pixels is futile for non-deterministic sensory data; predict in representation space instead.
- Techniques like Barlow Twins, VICReg and SI-Gaussian help avoid collapse and learn rich representations.