Key Points
- LeCun left Meta because his work on world models had limited relevance to the company.
- He promotes the “neolab” model—start‑ups focused on fundamental AI research.
- Examples cited include Mira Murati’s “Thinking Machines” and Ilya Sutskever’s “Safe Superintelligence”.
- His new AI architecture learns physics from video and uses past experiences as “emotions”.
- A simple pinching example illustrates how emotions guide future predictions.
- LeCun predicts early, limited versions within twelve months and larger systems in a few years.
- He sees more intelligence as a way to reduce human suffering and improve decision‑making.
- The interview lasted three and a half hours, ending with them as the only diners.
Why LeCun Walked Away From Meta
Yann LeCun said one driver for his departure was that his research on world models and artificial‑general‑intelligence‑type projects showed potential in areas such as jet engines and heavy industry—areas Meta was not interested in. He found investors willing to back the next generation of AI technologies.
The “Neolab” Vision
LeCun describes a “neolab” as a start‑up that does fundamental research, which he believes is the most fertile ground today. He cites Mira Murati’s “Thinking Machines” and Ilya Sutskever’s “Safe Superintelligence” as examples of work that challenges investors to understand advanced AI concepts.
A New Architecture Grounded in Physics and Experience
His latest model uses video input to give AI an understanding of the physical world, allowing it to make better predictions about what will happen next. The model also relies on “emotions,” meaning past experiences and evaluations, to guide its predictions. LeCun illustrated this with a simple example: if someone pinches you, you feel pain, and the mental model of the pincher changes, leading you to recoil the next time—an anticipation driven by fear or avoidance.
Timeline for Early Deployments
LeCun expects “baby” versions of this technology within twelve months and larger‑scale systems within a few years. He cautions that it is not yet superintelligence but a path toward it, acknowledging that unknown obstacles may exist while expressing optimism.
Legacy and Outlook
When asked about his legacy, LeCun said increasing the amount of intelligence in the world is his goal. He believes more intelligence means less human suffering, more rational decisions, and a deeper understanding of the universe, adding succinctly, “We suffer from stupidity.” The interview lasted three and a half hours, and by the end they were the only customers left in the restaurant.
Source: arstechnica.com