Yann LeCun on Objective-Driven AI and Why LLMs Won't Get Us to AGI

Sam Charrington and Yann LeCun on JEPA, world models, why autoregressive LLMs are a dead-end for AGI, and Meta's open-source strategy.

Meta's top AI scientist explains why he thinks today's chatbots are impressive but missing something important, and how his lab is trying to build AI that really understands the world.

1. 1JEPA-style world models are more likely to reach human-like understanding than pure LLMs.
2. 2Open-source foundation models raise the floor for the whole ecosystem.
3. 3Video prediction, not text prediction, is the next big bet.
4. 4Planning + hierarchical reasoning are missing from today's chatbots.
5. 5Regulation should target applications, not research.

• company
  Meta
• research
  V-JEPA